F? .:
E... a?

5:1, ,9?

wt "Hm.

I
. a
41 .43 .(v :v...‘

.umfiwx.

 

 

 

 

 

This is to certify that the
thesis entitled

CONSONANTAL AND SYLLABIC ADAPTATIONS
IN ENGLISH LOANWORDS IN MANDARIN

presented by

Li-jen Shih

has been accepted towards fulfillment
of the requirements for the

MA. degree in Linguistics

 

 

@AL

 

Maj?» Professor’s Signature

May 10, 2004

 

Date

MSU is an Affirmative Action/Equal Opportunity Institution

.0-0-o-.-.-.-o-n—l-o-o-o-o-o-O-o-I--O-O-O-o-o-o-a-o-I-o--o—.-.-o-.--.-o-o-o -<.-‘-I”.-‘-I-— - A

 

 

 

 

 

LIBRARY
Michigan State
University

 

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

MAR 0 9 2015

 

 

 

 

 

 

 

 

 

 

 

6/01 cJClRC/DateDue.p65—p.15

CONSONANTAL AND SYLLABIC ADAPTATIONS
IN ENGLISH LOANWORDS IN MANDARIN

By

Li-jen Shih

A THESIS

Submitted to
Michigan State University
In partial fulfillment of the requirements
For the degree of

MASTER OF ARTS
Department of Linguistics and Germanic, Slavic, Asian and African Languages

2004

ABSTRACT

CON SONANTAL AND SYLLABIC ADAPT ATIONS
IN ENGLISH LOANWORDS IN MANDARIN

By

Li-jen Shih

This thesis examines consonantal and syllabic adaptations that English forms
undergo when borrowed into Mandarin Chinese (Standard Chinese). I argue that English
loanword input to Mandarin is best treated as acoustic signals without any phonological
structures. My argument is based on the observation that English non-contrastive features
may appear in the output of the loanword adaptation and that two English segments in an
onset cluster may merge into one.

I also argue that when English forms enter Mandarin, the acoustic signals are
interpreted using the allophones of Mandarin rather than the phonemes. This fact
suggests that although the Operative Level of the loanword adaptation and the lexical
level of Mandarin have the same function, their inputs are composed of segments from
different sources.

Finally, I argue that the preservation-deletion asymmetry found in the coda liquids
is attributed to the difference in salience. I suggest that though liquids are unsalient in
nature, English coda liquids that are preceded by front vowels are perceived as salient

and thus must be preserved in Mandarin loanword adaptation.

TABLE OF CONTENTS

CHAPTER 1
INTRODUCTION .............................................................................................................. 1
1.1 Outline .................................................................................................................... 1
1.2 Literature Review .................................................................................................... 6
1.2.1 Silverman (1992) ........................................................................................... 6
1.2.2 Yip (1993) .................................................................................................... 10
1.2.3 Paradis & Lacharite’ (1997) .......................................................................... 12
1.3 Main Issues ........................................................................................................... 16
1.4 Optimality Theory: Markedness, Faithfullness and Constraint Interaction .......... 18
1.5 The Mandarin Sound System ................................................................................ 20
1.5.1 Consonants ................................................................................................... 20
1.5.2 Vowels ......................................................................................................... 23
1.5.3 Syllable Structure ......................................................................................... 25
CHAPTER 2
DATA AND GENERALIZATION .................................................................................. 30
2.1 Unchanged Loanwords ......................................................................................... 30
2.2 Changed Loanwords ............................................................................................. 30
2.2.1 Feature Change in Licit Consonants ............................................................. 30
2.2.2 Feature Change in Illicit Consonants ............................................................ 35
2.2.3 Adaptation of Illicit Syllable Structures ...................................................... 38
CHAPTER 3
ANALYSES ...................................................................................................................... 42
3.1 The nature of Loanword Input .............................................................................. 42
3.1.1 Aspiration in English Stops and Affricates .................................................. 43
3.1.2 Lip Rounding in English [1] .......................................................................... 46
3.1.3 Syllabicity in English [1] .............................................................................. 47
3.1.4 Velarization of English Coda [l] .................................................................. 49
3.1.5 Nasalization of English Vowels Before Nasals ........................................... 51
3.1.6 Segment Merger ........................................................................................... 53
3.1.7 Summary ...................................................................................................... 56
3.2 At the Perceptual Level ......................................................................................... 56
3.2.1 Licit Consonants .......................................................................................... 56
3.2.2 Illicit Consonants ......................................................................................... 59
3.3 Mandarin Palatals .................................................................................................. 61
3.4 Character Selection ............................................................................................... 68
3.5 At the Operative Level .......................................................................................... 75
3.5.1 Mandarin Syllable Structure Constraints ..................................................... 76
3.5.2 Vowel Epenthesis ......................................................................................... 76

iii

3.5.3 Liquids in the Coda Position ........................................................................ 79

3.5.4 [m] in the Coda Position .............................................................................. 95

3.5.4.1 Polysyllabic Input ............................................................................... 95

3.5.4.2 Monosyllabic Input ............................................................................. 98

3.5.4.3 An Attempt to Account for Cases such as Lynn ............................... 104
CHAPTER 3

CONCLUSION ............................................................................................................... 1 07

REFERENCES ............................................................................................................... 109

iv

CHAPTER 1

INTRODUCTION

1.1 Outline

This thesis examines consonantal and syllabic adaptations that English forms
undergo when borrowed into Standard Chinese, which I refer to as Mandarin
throughout this thesis simply for convenience. These two types of adaptations can be
seen in the borrowing of the proper name Louise, Louise —) [lu.i.si], where [z] is
adapted to [s] by devoicing, and [i] is epenthesized at the end of the form to create a
new syllable.l Clearly, the consonantal adaptation is triggered by Mandarin segment
inventory constraints, which disallow voiced obstruents, and the syllabic adaptation is
triggered by the phonotactic constraints, which only permit [n] and [13] in the coda
position of a syllable. The adaptation will be analyzed within the framework of
Optimality Theory (Prince & Smolensky 1993) in this thesis.

The data are primarily drawn from common English proper names listed in the
appendix of the Oxford Advanced Leamer’s Dictionary of Current English (1984),
supplemented with those collected myself. Since tones are irrelevant to the analyses,

they are omitted.

 

' Coda liquids in the English forms sometimes are round to be deleted in the adaptation, e. g., Mark —>
[makha]. Details with respect to coda liquid deletion are provided in §3.5.3.

Three things need to be noted here. First, most borrowings through transliteration

in Mandarin are proper names. The way Mandarin deals with non-proper—name new

words from foreign sources is usually to look at how the objects represented by these

words function. ‘Computer’ is a good example:

(1)

[tjan]2 ‘electricity’
[naw] ‘brain’
[tjannaw] ‘Computer’

Since most foreign words that enter Mandarin through transliteration are proper

names, they serve as the bulk of the data in this thesis.

Second, I simply assume that American English is the main source of the data

because of the great American influence on the media. However, it is possible that

some of the loanwords are based on British English due to the domination of British

English in East Asia in the past. Whether we are dealing with British English or

American English as the source language makes a slight difference in the analysis.

For example, it is known that British English replaces post-vocalic [r] with a schwa or

simply deletes it. If the loanwords are based on British English, then post-vocalic [r]

 

2 In this thesis 1 adopt Duanmu’s (2000) proposal that at the surface level, a syllable-initial consonant
and a pre-nuclear glide merge into a single segment in a way that the glide palatalizes or adds a
secondary articulation to the preceding consonant. Details about Mandarin syllable structure will be
shown in §1.4.

may not be in the input.

Third, Chinese characters influence the loanword adaptation to some degree.

Every Chinese character corresponds to a syllable and possesses a meaning. Therefore,

when a loanword enters Mandarin, Mandarin speakers prefer to have a Mandarin form

that is composed of characters with positive meanings. At the initial stage of

borrowing, a loanword entry may have several Mandarin forms. At a latter stage, a

particular form is chosen and enters dictionary. This standardized form may not be as

phonetically close to the input as the others. However, as we will see in §3.4, there is

a limit to how far the form could be from the input. Otherwise, we would find a

Mandarin form that is entirely different from the input. Our data show that only a few

types of feature changes are made when the character selection factor exerts its

influence. The interaction of the character selection with the faithfulness constraint

reveals that some feature changes make more difference in perception than others.

Silverman (1992) investigates English loanwords in Cantonese and proposes that

'loanwords do not come equipped with their own phonological representation...the

input to loanword phonology is merely a superficial non-linguistic acoustic signal’

(p.289). He also suggests that in order for the incoming acoustic signals to surface in

accordance with the borrowing language’s phonotactic constraints, they must go

through two separate, yet ordered levels: the Perceptual Level and the Operative

Level. At the first Level, the acoustic signals are replaced with the native feature
matrices that are most approximate to their properties. The output of this level is the
input of the second level, at which some necessary phonological processes apply, such
as epenthesis and deletion. These two vieWpoints are maintained by Yip ( 1993), but
she argues that no rules are involved in the course of loanword adaptation and
reanalyzes the data in the framework of the Optimality Theory. Paradis & Lacharité
(1997) take an opposite position with respect to the nature of loanword input. They
claim that ‘often, loanwords enter the borrowing language (L1) with structures (that is,
segments and sequences) that are, from the point of view of L1, ill-formed’ and that
‘most of our results are incompatible with phonetic approximation views (for
example, Silverman 1992, Yip 1993) (p.380). Although they agree that loanword input
is the phonological output of the lending language, they admit that it is not clear
whether the input is the output of L1 at the lexical or postlexical level.

In this thesis, I will show that when Mandarin borrows words from English, it is
the output of the postlexical level of English that forms the basis of the loanword
input. The evidence is mainly drawn from the observation that English
non-contrastive features such as aspiration in voiceless obstruents may appear in the
Mandarin corresponding segments. Furthermore, I will argue that English loanword

input enters Mandarin without structures. This is to say that the input is best viewed as

superficial acoustic signals. I will support this claim by showing that two English

segments may merge into one in the course of the adaptation.

In this thesis, I will also argue that the loanword adaptation not only operates on

the output of the postlexical level of English, but also operates on the output of the

postlexical level of Mandarin. I will show that the acoustic signals are interpreted as

phonological representations using Mandarin allophones, rather than using its

phonemes.

Finally, I will show in this thesis that while vowel epenthesis is very much

preferred to consonant deletion as a strategy to repair ill-formed syllable structures

(e.g., Louise —> [lu.i.si] *[lu.i], Qreg —-> [ke.lej.ke] *[lej.k9]). liquid codas tend to

delete when preceded by back vowels (e.g., Bugbara —-) [pa.pa.|a] *[pa.er.pa.la]). I

suggest that this fact is due to their relative unsalience as compared to those liquids

preceded by front vowels and other illicit coda consonants such as obstruents.

The remaining sections of this chapter are organized as follows. §1.2 is a

literature review, which includes Silverman (I992), Yip (1993) and Paradis &

Lacharité (1997). §1.3 is a brief overview of Optimality Theory. §1.4 introduces the

Mandarin sound inventory and syllable structure. In Chapter 2, I will present the data

and point out their main characteristics. More specifically, §2.1 comments on

unchanged loanwords, and §2.2 discusses loanwords that undergo adaptation. In

Chapter 3, I will present my analyses, which are further divided into three sections: In

§3.1 I provide evidence for Silverman’s view on the nature of the input; §3.2

discusses the operations regarding feature changes at the Perceptual Level; §3.3

discusses the phonological processes at the Operative Level. In Chapter 4, I present

the conclusion.

1.2 Literature Review

1.2.1 Silverman (1992)

Silverman investigates English loanwords in Cantonese and gives several

suggestions. First, he argues that loanwords do not enter the borrowing language with

their own phonological representation; instead, loanword input consists of a sequence

of non-linguistic acoustic signals. Behind this view is the assumption that speakers of

the borrowing language have no access to the phonological system of the lending

language.

The second suggestion Silverman makes is that loanword phonology contains

two distinct, ordered levels. The first level, termed the Perceptual Level, is a stage at

which the acoustic signals are parsed into segment-sized chunks for which the native

feature matrices that are closest to their articulatory and/or acoustic properties are

provided. In addition to feature matrices, prosodic representation such as syllable

(i

nodes and, in the case of Cantonese, a binary foot template is provided at this level as

well. The processes at this level are constrained by the borrowing language’s

phonological system. It is not until the derivation reaches the second level, the

Operative Level, that the raw materials undergo real phonological processes. At the

second level, the borrowing language’s phonotactic constraints hold and trigger

various phonological operations, like epenthesis and deletion, so that the preliminarily

processed materials can surface in conformity with the native syllable and metrical

structure constraints. To exemplify the distinction of the Perceptual Level from the

Operative Level, Silverman refers to the deletion of post-consonantal liquids:

(2) (p.290)
English Cantonese
a. bgeak —> [piklik]
pgint —> [philin]
cgeam -—> [keylim]
b. pzinter —> [phentha]
bgoker —> [puk.kha]
fLeezer —> [fisa]

In (2), all the English forms begin with a stop-liquid cluster. In the Cantonese forms,
however, only those in (2a) retain the liquid. One hypothesis is that Cantonese
speakers only perceive the liquids in (2a) but not those in (2b). This hypothesis cannot

be supported because a near-minimal pair is found, print —> [phi.lin] vs. printer —)

[phentha]. It is unlikely that Cantonese speakers perceive the liquid in print but not the
one in printer. Silverman suggests that the liquids in (2a) are retained because the
retention makes the output forms bisyllabic. In (2b), on the other hand, since the
retention of the liquids would make the output forms exceed bisyllabicity, they are
deleted, and the resulting output forms are, again, bisyllabic. The main point here is
that the liquids in both (2a) and (2b) must be perceived first, and it is at the second
level, where Cantonese phonotactic constraints hold, that the decision of whether they
should be retained or deleted is made.

The third suggestion is that at the Perceptual Level, syllable nodes are provided
for the most salient segments. The segments perceived as syllabic include vocalic
sonority peaks and phonetically salient consonants such as [s]. Silverman claims that
the salience of [s] is due to its duration and sibilance. When [5] occurs post-vocalically,
either pre-consonantal or form-final, it is perceived as syllabic and provided with a

syllable node. Some derivations are shown ill (3):

(3) (13-319)

Input printer print tips file
Scansion One (I) (I) (I) (I)
/\ l /\ |
0 O 0 O 0 U
| | | | l l
[phl e n t" a] [phl i n] [th i ph s] [fajl]
Scansion Two (I) (I) (I)

>9
>

/\ /\
o o o 0 0 a

0 0
/i\ /l /\ A /l\ A /l\ /l\
[phen tha] [phi lin] [thiphsi] [faj low]

At the Perceptual Level (Scansion One), the acoustic signals are parsed into
segment-sized chunks for which the native feature matrices that most approximate
their properties are provided. In addition, the vocalic sonority peaks and the most
phonetically salient consonants are assigned syllable nodes, which associate with a
binary foot template. At the Operative Level (Scansion Two), Cantonese phonotactic
constraints hold and real phonological processes such as epenthesis and deletion apply.
Thus, the liquid in printer gets deleted, while those in print and file are retained
through vowel epentheses. The phonetically salient consonant [s] in tips, as it appears
form-finally, is perceived as syllabic at the Perceptual Level and thus is retained by
epenthesizing [i] after it at the Operative Level.

The fourth suggestion is that Cantonese loanword phonology exists as a separate

grammar from the native phonology. This suggestion is based on the observation that

9

those phonological processes applying at the Operative Level are hardly found in the

native phonology. For example, Cantonese native phonology involves no such

processes as epenthesis, deletion or resyllabification, while those processes are clearly

found at the Operative Level of Cantonese loanword phonology.

1.2.2 Yip (1993)

Like Silverman, Yip examines English loanwords in Cantonese. Though she

adopts Silverman’s claims that some of the contrasts existing in English such as

voicing are never perceived by Cantonese speakers and that the processes loanwords

undergo are divided into two levels, she argues that no rules are involved and the data

can be accounted for within the framework of Optimality Theory. From OT’s

perspective, loanwords are those words that are borrowed from a language possessing

a set of ranked well-formedness constraints into a language possessing a different set

of ranked well-formedness constraints. As borrowing proceeds, some adjustments

regarding segmental and/or prosodic structures may be made to meet the new set of

constraints. These adjustments are minimal: in other words, the output forrns are made

as close to the input forms as possible as long as the new set of constraints can be

satisfied. Yip claims that the most crucial constraints in Cantonese loanword

phonology are: (i) a set of syllable structure constraints, (ii) the preference for

bisyllabic words, and (iii) the force that makes the output as close to the input as
possible.

It is suggested in Silverman (1992) that the most phonetically salient segments of
the input are provided with syllable nodes. Yip notices that although this suggestion
succeeds in accounting for the fact that [s] is consistently preserved while liquids are
not, it fails to explain why it is always the stops rather than the liquids that are
preserved in pre-vocalic stop-liquid clusters (e.g., printer —-9 [phen.tha]), given that
neither is provided with syllable nodes. By showing evidence found in speech errors,
she argues that liquids are less salient than stops and this relative unsalience makes
liquids more vulnerable to deletion unless other factors such as the pressure of
bisyllabicity are involved.

Another argument Yip makes is that loanword phonology should not be viewed
as a separate grammar from the native phonology. She shows that the phonological
processes found at the Operative Level actually can be motivated in the native
phonology. In order to explain why some phonological processes observed in the
loanword phonology are never found in the native phonology, Yip suggests that the
input for which such processes are needed are never present in the native phonology.
To illustrate the motivation of epenthesis in Cantonese native phonology, Ylp refers to

augmentation of kinship names: they are usually monosyllabic but can be augmented

by adding the prefix [a], given that they are subject to a bisyllabic minimum. For

example, Yip can be augmented as [ajip].

1.2.3 Paradis & Lacharité (1997)

The loanword analyses done by Paradis & Lacharité (1997) are based on a

statistical investigation of segmental and syllabic malformations found in a corpus of

545 French loanwords in Fula. They observe that the illicit segments are preserved via

minimal adaptation in a vast majority of cases but are deleted when contained in

prohibited syllable structures. Paradis & Lacharité come to the conclusion that input

information is maximally retained in the output unless preserving it would violate

what they call the Threshold Principle, by which the maximal number of steps needed

to repair an illicit structure within a constraint domain is set at two for all languages.

The examples in (4) and (5) briefly illustrate their points:

(4) Three ways of adapting illicit /v/ (p.400)
/v/ —> [w] (62/81 cases of adaptation — 76.5%)

French Fula Gloss
avocat /ayoka/ —) awoka ‘lawyer’
civil /siyil/ -—> siwil ‘civil’

/v/ —) [b] (14/81 cases of adaptation — 17.3%)

livre /liyr/ —> lixbalr ‘book’
vinaigre lyinegr/ —> binegara ‘vinegar’
/v/ —> [f] (5/81 cases of adaptation — 6.2%)

mouvement /muy_ma/ ——> mufmal] ‘movement’

télévision /teleyizjo/ —> telefisjalj ‘television’

As shown in (4), illicit /v/ can be adapted in three ways: (i) inserting [+sonorant]

(/v/ ——> [w]), (ii) delinking [+continuant] (/v/ —-> [b]). and (iii) delinking [+voice] (/v/

—> [1]). Note that the repair strategy in (i) is much more frequently employed than

those in (ii) and (iii) (the occurrences of /v/ ——> [w] make up more than three-fourths of

the instances). The reason for this striking gap is that in /v/ -> [w] all the information

contained in /v/ is preserved (/v/ is adapted by adding [+sonorant]), but in both /v/ —)

[b] and /v/ —> [f] some information in /v/ is lost.

(5) Two ways of fixing an ill-formed syllable
a. Vowel epenthesis (p.406, 407)

French Fula Gloss
drapeau lgapo/ —> d_arapo ‘flag’
boisson /p_w_as‘5/ —> b_u_w_asorj ‘drink’
course /ku;§/ —> kur_s_i ‘course’

b. Consonant deletion (p.408)

voyou Aflaj u/ —> waju ‘burn’
cuivre /kgiy_r/ —> kigi ‘copper’
biscuit /bisk_qi/ —> biski ‘biscuit’

Fula does not allow complex onsets/codas. To fix this structure, either a vowel is

epenthesized to break the consonant cluster (5a) or one of the consonants is deleted

(5b). Vowel epenthesis is considered a two-step process (nucleus insertion plus vowel

spreading, e.g. lgapo/ —9 glgrapo, or glide spreading, e.g. /l_)_was5/ —) l_)_u_v_v_ason),

satisfying the Threshold Principle. It is only when an illicit consonant is contained in

complex onset/coda that consonant deletion takes place. In this situation, repair by

vowel epenthesis requires three steps (two for epenthesis of a vowel and one for

adaptation of the illicit consonant), going beyond the limit set by the Threshold

Principle. As a result, the illicit consonant undergoes deletion, which takes only one

step.

Unlike Silverman, Paradis & Lacharité claim that loanwords enter the borrowing

language with structures. They argue that ‘the input is immediately interpreted as a

phonological representation by L1 (the borrowing language) and handled by its

14

constraint set’ (p.380).

Following Yip (1993), Paradis & Lacharité reject Silverman’s view that

loanword phonology constitutes a separate grammar from the native phonology. They

suggest that regarding loanword adaptation processes as rules results in the separate

grammar view. If phonological processes are considered to be reactions to constraint

violations, then the fact that in Cantonese some processes are specific to loanword

adaptation can be explained: native words never violate the constraints that are

responsible for such processes.

The schema in (6) represents Paradis & Lacharité’s loanword model:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

(6) (p.394)
The lending language The borrowing language
Dictionary Dictionary
Phonological Phonological
Constraints: Constraints:
Lexical and Lexical and
Postlexical Postlexical
Levels Levels
Phonetic
Output

 

 

 

Paradis & Lacharité’s loanword model differs from Silverman’s in two

fundamental aspects. The first is with respect to the nature of the input. In Silverman’s
model, the input consists of acoustic signals, while in Paradis & Lacharité’s, the input
contains phonological representations of the lending language, although they do not
state which level. The second aspect is regarding whether there is such a thing as
loanword phonology. Silverman argues for its existence by stating that the processes
applying at the Operative Level of Cantonese loanword phonology are hardly
observed in the native phonology. Following Yip, Paradis & Lacharité suggest that the
absence of such processes in Cantonese native phonology is due to the lack of input

for which such processes are needed.

1.3 Main Issues

In §3.1, I will show that when English forms are borrowed into Mandarin, the
input is the output of the postlexical level of English. Evidence is obtained from the
observation that many English allophonic features appear in the Mandarin forms,
which forces us to assume the presence of those features in the input. An example is a
Mandarin syllable which corresponds to English syllable-final [1]. Our data show that
English syllable-final U] is matched to Mandain [9i] or [9y], e.g., Bush —> [pa.gi],
Cash —> [khajcy] ([si] is another Mandarin syllable to which English syllable-final [j]

is matched, but it is irrelevant to the present discussion). To account for the emergence

16

of [labial] in [9y], it seems necessary to assume that this particular feature is present
in the input (note that English [1] contains the non-contrastive feature [labial]), and for
some reason it is deleted in Bush —> [pugi]. If one analyzes it the other way around by
assuming that this feature is not present in the input but inserted in Cash —) [khajcy],
then one has to explain where this feature is from. Unfortunately, this does not seem
to be possible.

Recall Silverman’s claim that at the Perceptual Level, the acoustic signals are
parsed into segment-sized chunks, and these chunks are provided with the native
segments that are closest to their acoustic/articulatory properties. It is not clear
whether ‘the native segments’ are referred to as the phonemes or the allophones of the
borrowing language in his article. Our data show that ‘the native segments’ Silverman
talks about have to be the allophones. I mention in the preceding paragraph that
English syllable-final U] may be matched to the Mandarin syllable [9i]. If it is a
Mandarin phoneme that fills in the chunk containing the acoustic signals of [j] at the
Perceptual Level, then the phoneme serving this work is probably [5] or [5], given that
they are closest to [j] in terms of acoustic/articulatory properties. [9] is not a choice
because it is an allophone in Mandarin (Lin 1989, Duanmu 2000). These two
segments in this particular position then are evaluated by Mandarin phonotactic

constraints at the Operative Level and judged as illicit structures (only [n] and [1]] are

allowed in the coda position). Due to their salient nature, they are retained by

epenthesizing an apical vowel after them (Lin 1992), yielding the output forms [si]

and [si]. In this way, [9i] can never be derived. If it is the allophone [c] that is

provided at the Perceptual Level, then we have no problem deriving the syllable [oi].

At the Operative Level, [9] in this particular position is judged as an illicit structure

and is fixed by vowel epenthesis. The inserted vowel obtains its quality from the

preceding [9] through assimilation, yielding the attested output form [9i]. In §3.2.2, I

will give a complete analysis arguing that at the Perceptual Level, the English

acoustic signals are replaced by Mandarin allophones rather than Mandarin

phonemes.

1.4 Optimality Theory: Markedness, Faithfullness and Constraint Interaction

OT (Prince & Smolensky 1993, Kager 1999) sees the grammar of a language as a

system of conflicting forces. These forces are universal and exist in the grammar in

the form of constraints. The forces or constraints can be divided into two basic groups

— markedness and faithfulness. Markedness constraints state that unmarked

structures are universally favored over marked structures. For example, an open

syllable is favored over a closed syllable (NO-CODA); only some types of consonants

can serve as the coda in many languages (CODA CONDITION); a voiceless obstruent is

18

favored over a voiced obstruent (VOICED OBSTRUENT PROHIBITION (VOP)), and so on.
Each faithfulness constraint states that the output must preserve the properties of the
input, which can be defined in terms of features, segments, or prosodic structures. For
example, the output segment must have the same value for the feature [voice] as the
input segment (IDENT-IO(voice)); segments in the output must have correspondents in
the input (DEF-IO); and the output must have the same linear order of the segments as
the input (LINEARITY—IO). Faithfulness to the input can be understood as pressure to
preserve lexical contrasts.

Markedness constraints and faithfulness constraints are inherently in conflict.
The English proper name Luke surfaces as [lu.khe], where [a] is epenthesized at the
Operative Level. The epenthesis of [9] indicates that the markedness constraint CODA
CONDITION is satisfied. However, the satisfaction of CODA CONDITION is
accomplished at the cost of violating the faithfulness constraint DEP-IO.

This example shows us that for two conflicting constraints, one has priority to be
satisfied over the other. This idea is built into OT in the form of hierarchical ranking,
in which one output candidate surfaces as the actual output because it incurs lesser
constraint violations than others.

The interactions of constraints can be represented by a tableau. In a tableau,

output candidates are listed vertically in a random order, while the constraints are

19

arranged horizontally in a rank-descending order from left to right. (7) shows the

tableau in which [lu.kha] is selected as the output.

 

 

 

 

(7)
Iukh ‘Luke’ CODA CONDITION DEP-IO
CE'a. lu.kho *(o)
b. hilth *(kh)l

 

 

 

‘9” marks the Optimal output candidate. The optimal output candidate is (7a) because

the other candidate violates the higher-ranked CODA CONDITION. A violation of a

constraint is marked by ‘*’; ‘1’ denotes that the violation of the particular constraint is

fatal; and the shading of a cell indicates the violation(s) of the corresponding

constraint is irrelevant to the result.

1.5 The Mandarin Sound System

1.5.1 Consonants

Mandarin has nineteen underlying consonants, which are listed in (8):

20

 

(8) (Lin 1989, Duanmu 2000)

labial dental retroflex velar

stop p t k
ph th kh
affricate ts [s
t Sh Lsh
fricative f s s x
nasal m n 13
liquid 1 r

Two things regarding (8) are noted here. First, Mandarin obstruents do not
contrast in voicing. They are unspecified for [voice] at the lexical level and obtain the
feature specification [—voiced] at the postlexical level by a UG default rule, in the
sense that minus is the unmarked value for [voice] in obstruents. In contrast, English
obstruents are distinguished in terms of voicing at the lexical level. As a result, when
English forms enter Mandarin, [+voiced] obstruents must be devoiced (in some cases,
[v] is adapted to [w] rather than [f]), as we will see in the data to be presented in
§2.2.2. Secondly, Mandarin stops and affricates contrast in aspiration at the lexical
level. Although English voiceless stops also contrast in aspiration, as the distinction
between the first segment in [th]0p and the second segment in s[t]0p, it is at the
postlexical level rather than at the lexical level that this contrast takes place. English
voiceless stops are unspecified for aspiration at the lexical level, and a
language-particular rule assigns them [+spread g1] at the postlexical level if they are

syllable-initial.

21

In addition to the nineteen consonants, Mandarin has the three palatals [t9], [tch]

and [9]. They are considered allophones rather than phonemes because of their

distributions: [t9 to“ 9] appear only before the high front vowels [i y] and the glides [j

(1] while the dental sibilants [ts tsh s], the retroflex sibilants [Ls Ls“ s] and the velars

obstruents [k kh x] never do. (9) shows that [tc toh c] are in complementary

distribution with [ts tsh s], [is Ls" 5] and [k kh x]:

(9) (Duanmu 2000: p.27)

 

 

 

 

 

 

 

 

 

[ts t9“ 9] [ts ts“ s1 [Ls is“ s] [k k“ xl
followed by [i y]/[i q] v x x x
not followed by [i yl/[i II] x \l ‘1 Ni

 

Lin (1989) suggests that [to tr]:h 9] surface by the phonemes assimilating to

[—back] of following [i y] or [i q]. This process is illustrated in (10) using Its tsh s/:

(10)

Underlying Representation

/ts tsh s/ +

/i y/ or /j q/

[—back]

__)

[t9 t9” 9]

+

Surface Representation
[i y] or [i II]

"a
u.‘.
..__
o.__
.....
a...
o
-o
.....
.....
o...
.._
O...
0.
-...
o...
a...
..._

In many studies, a syllable-initial consonant and a following pre-nuclear glide

are considered two independent segments. However, Duanmu (2000: §4) proposes

22

that at the surface level the two segments merge into one, which is achieved via the
glide palatalizing or adding a secondary articulation to the syllable-initial consonant.
In the former case, the results are [to tr;h g] or [th tgwh cw], depending on which glide

(U] or [q]) is involved. (11) shows the palatalization of [ts tsh s] by a following U] or

[II]:
(1 l )
Underlying Representation Surface
Representation
/ts tsh s/ + /j/ —> [t9 t9“ 9]
/ts tsh s/ + AV —-> [t9w t9“ 9“]

Notice the difference in the treatment of the glides between (10) and (11). In (10),
they are independent segments at the phonetic level, while in (11) they merge with the
preceding consonant. Duanmu’s proposal is adopted in this thesis. In the discussion of

Mandarin syllable structure, this issue will be elaborated in more detail.

1.5.2 Vowels

Mandarin has five underlying vowels, which are listed in (12):

23

(12) (Lin 1989, Duanmu 2000)

high i y u
mid a
low a

Both the low vowel [a] and the mid vowel [a] have several surface
representations. However, agreement has not been reached among scholars with
respect to how they should be transcribed. One thing that is clear is that within a
syllable a preceding glide or a following segment may affect the two vowels’
backness, but not their height. For instance, according to S. Xu (1980: p.184), the mid
vowel [9] surfaces as [E] if preceded by [j] (/jo/ —> [jE] ‘leaf ’) ([E] denotes a vowel
that is higher than [8] but lower than [e]) but as [o] if preceded by [w] (/we/ —> [wo]
‘1’). Lin (1989: p.51—52) claims that [a] surfaces as [a] if followed by [w] or [g] and
that [3] always becomes [e] if adjacent to a high front glide and [o] if adjacent to a
back glide. In this thesis, [e a o] and [a] represent the surface forms of the mid vowel
and low vowel, respectively.

In addition to the five underlying vowels, Mandarin has two so-called apical
vowels: the dental apical vowel, which only occurs after the dental sibilants [ts tsh s],
and the retroflex apical vowel, which only occurs after the retroflex sibilants [[p [sh s]
and [r]. In this thesis, the high back unrounded vowel [i] is used to cover both types of

apical vowels.

24

The status of the apical vowels is still a debated issue. In C. Cheng 1973, the
apical vowels are viewed as deriving from some underlying vowels. Other scholars,
on the other hand, suggest that the apical vowels should be treated as syllabic
consonants [z] and [r], given that the properties the apical vowels possess usually
belong to consonants — syllabic [z] is similar to voiced [s] and syllabic [r] is similar
to pre-nuclear [r] (E. Pulleyblank 1984, Fu 1986). Whether the apical segments in
question are real vowels or should be considered syllabic consonants is not the
concern of this thesis. Based on the fact that the distributions of the apical vowels are
limited and predictable, the claim made by Lin 1992 that they are not underlyingly
present and can be regarded as epenthetic vowels is adopted. (13) shows that the two
Mandarin morphemes ‘four’ and ‘ten’ lack vowels in their underlying representations,
and it is at a later point in the derivation that nuclei are epenthesized and assimilate to

the preceding consonants. (For more detail, see Lin 1992:236—241)

(13) (Lin 1992:238) (N denotes nucleus)

UR SR
/s/ —> SN —-> [si] ‘four’
/§/ —> sN -—> [5i] ‘ten’

1.5.3 Syllable Structure

In the traditional view, a Mandarin syllable is (C)(G)V(X), where C is a

25

..
F“

 

.m—v'

 

 

 

 

 

consonant, G a glide, V a vowel and X a nasal or a glide. The length ranges from a
maximum of four segments to a minimum of one segment. The diagram in (14)

illustrates the structure of a Mandarin syllable in the traditional view:

( 14)
S S: Syllable
/\ 1: Initial (Onset)
I F F: Final
/\ M: Medial
R R: Rime
/\ N: Nucleus
E: Ending (Coda)

The major characteristic of this structure is the presence of the Final node, which

dominates the pre—nuclear glide and the rime. Several proposals have been made with

respect to how a Mandarin syllable should be represented. In Lin (1989), the

traditional view of Initial/Final division is maintained while a post-nuclear glide is

syllabified as the second half of the nucleus rather than as the coda. In Bao (1996), a

pre-nuclear glide is considered to be part of the onset and thus the traditional

Initial/Final division is not maintained. Duanmu (2000: §4) discards the traditional

way of analyzing a Mandarin syllable and suggests that Mandarin has only two types

of syllables — the full syllable and the weak syllable. The full syllable has three

26

timing slots and is heavy, and the weak syllable has two timing slots and is light. In

Mandarin, lexical words in most cases are full syllables, while grammatical words are

always weak syllables. Since our data are all lexical words, only the full syllable is

considered in this thesis. The structure of the full syllable is shown in (15):

(15) (p.79)
3
/\
O R
/\

One of the main differences between the traditional representation as shown in

(14) and Duanmu’s analysis is regarding the status of the pre-nuclear glide. In the

traditional representation, a pre-nuclear glide is always an independent segment, no

matter whether or not it is preceded by another consonant. In Duanmu’s analysis, a

pre-nuclear glide is an independent sound only when it is syllable-initial. If a

preceding consonant appears, the pre-nuclear glide palatalizes or adds a secondary

articulation to the consonant, and the result is that both segments merge. Duanmu

states that this is because a pre-nuclear glide is syllabified as part of the onset and

there is only one timing slot for the onset. To illustrate this idea, Duanmu refers to an

27

example originally pointed out by Chao (1934) in which /s/ and /w/ in the onset merge

into a single segment at the surface level:

(16) (Duanmu 2000: p.86)

Underlying Representation Surface Representation
S S
/\ /\
O R O R
l /\ l /\
l m m | m m
| | | | | l
X X X X X X
/\ | |
s u a n s u a n
Isuan/ [swan]

At the underlying level, this Mandarin morpheme contains four segments. Given that
a pre-nuclear glide is syllabified to the onset and that there is only one timing slot
available for the onset at the surface level, the syllable-initial segment /s/ and the
following pre-nuclear glide /w/ have to share the slot and merge to [sw].

One of the arguments Duanmu provides for a syllable-initial consonant and a
following pre-nuclear glide merging to a single sound at the surface level is the
obvious phonetic difference between English swan [swan] and Mandarin swan [swan]
‘sour’ (Chao 1934: p.42, Duanmu 2000: p.28, 86). Both forms have four segments at

the underlying level. However, the latter by no means sounds like the former, and the

28

difference between the two apparently falls on the part of sw: it sounds like one
segment followed by another in English but a single segment in Mandarin.

Duanmu’s view (the structure shown in (15)) is supported by the data of this
thesis. For example, when the English word [[0] ‘show’ enters Mandarin, the
Mandarin output is transcribed as [cjow] in the traditional view but as [cow] in
Duanmu’s view. Because loanword adaptation must be minimal, Duanmu’s view is
confirmed given that it costs more for [j] to be adapted to two segments ([cj]) than to
one segment ([9]). I will talk more about this issue in §3.2.2. In the following chapters,

all the Mandarin forms will be transcribed based on Duanmu’s view.

29

CHAPTER 2

DATA AND GEN ERALIZATION

2.1 Unchanged Loanwords

As the examples in (17) show, if the consonants and syllable structures involved

in the incoming English forms satisfy the requirements imposed by Mandarin (the

segmental inventory constraints and phonotactics), they surface without any

adjustment. This confirms the idea that in loanword adaptation the output is matched

to the input as closely as possible.

( l 7)

pie —> [phaj] Lynn
mufi‘in —> [mafan] Colin
Simon -—> [saj .man]3 [an
VVIIIy —> [wej.li] Tina
2.2 Changed Loanwords

2.2.1 Feature Change in Licit Consonants

1111

[lin]
[khe.lin]
[i.en]
[thi.na]

If the consonants in the English forms go through feature change, we naturally

suspect that they are illicit from the perspective of Mandarin and therefore need to be

 

3 It is possible that an English loanword has more than one Mandarin form. For example, an alternative

Mandarin form for Simon is [saj.man].

30

adjusted. In other words, given that the output is matched to the input as closely as
possible, there seems to be no reason to adjust the consonants in the input that can
surface. However, we find a small number of cases in which the input consonants
undergo feature change even though they are licit segments in Mandarin. Consider the

following examples:

<18) Ip" t“ 18‘] —> lp“ t“ khl

a. Beg —) [p_hej.ka]
b. [at -—> [fiajta]
c. Kevin —> [flajwan]

(19) [ph th kh] —-> [p t k]

a. flaula —> [p_o.1a]
b. [any -—) [togni]
c. Kennedy —> [kan.naj.ti]

(20) [t k] ——) [t k]

3. Steven ——> [§i.ti.fu]
b. Stanley —-> [§i.tan.li]
c. Sgotland —-> [su.ka.lan]

(21) [t k] —-> [th kh]

a. Steve —-> [sifiifu]
b. Stanley —> [sifianli]
c. Sgott —-) [silfiawthe]

In (18) and (19), the English forms begin with an aspirated voiceless stop. If we

31

compare the Mandarin forrns in (18) with those in (19), we find that only those in (18)

keep the feature [+spread g1]. A similar situation is observed in the comparison of (20)

and (21). In (20), [—spread g1] in the English voiceless stops occurring after [3] is

retained in the adaptation, while in (21), this feature becomes [+spread g1].

Another example showing that a licit consonant undergoes feature change is the

adaptation of English onset [r]. In (22), the English onset [r] surfaces without change,

whereas in (23), it is adapted to [1]. Note that some English forms have two Mandarin

equivalents.

(22) [f] —> [r]

a.

Redford —. [[wejfu] Shgek —+ [sirwejkhe]

flachel —> [[wejtchow]

b.

Ricky —-> [rwejtchi] Jeflgey —> [tcejforwej]
giro —> [[wejtha] E_r_ic —> [aj.rwej.k”a]

Hillary —> [cilarwej]

32

(23) [r] -> [1]

3.

Rachel —) [1ej.tchow] Rex —-> [lej.khe.si]
Raymond —~> [lejmalj] Reynolds —-> [lej.nwo.tsi]
Reagan —> [lej .kan] Ggeg —> [ke.lej.ke]
b.

Richard —) [litsha] H i 1 100’ —> [91-13-11]
Ricky —> [Iichi] Jeflzey —> [tcej.fo.1i]
Rita —) [lithal Egic —> [aj.li.lt“a]
c.

Robin —9 [1a. pin] Doga -—> [two.1a]
Roy —) [lwoj] Rudolf —~> flu.taw.fu]
Rice —~) [laj .si] Rose —> [1wo. si]

Several points are noted here. First, if [r] is not adapted to [1], it surfaces as [rw].
Secondly, it seems that the English syllable [rl] can be adapted to either [rwej] or [li].
In the comparison of (22b) with (23b), we find that only Richard has one Mandarin
equivalent. Thirdly, the English syllables [rej] and [re] seem to have two Mandarin
equivalents as well, one [rwej] and the other [lej] (22a, 23a). Finally, input [r] that is
followed by a vowel other than [I], [e] or [ej] is always adapted to [1] (23C).

The adaptation of coda [n] to [1]] also shows that a licit consonant may undergo

feature change. Compare (24) with (25):

(24) [n] —-> [n] (in the coda position)

Robig —> [1a.pig] Vlgceat —> [wagseg]
Viviaa —> [wej.wejam Susan. —) [su.§am
S imogg —> [saj .mag] Jonathan —) [tchaljnasam

33

(25) [n] -—) [g] (in the coda position)
Raymogd —> [lej.mag] Vincer_zt —-> [wenseg]
Jonathan —> [tchannasag] Damog —> [taj.meg]

Take Simon from (24) and Damon from (25) for example. In the adaptation of Simon,
all the segments in the second syllable surface intact. In contrast, exactly the same
syllable in Damon surfaces with the coda [n] changed to [1]]. This seems to show that
[n] and [13] are interchangeable in the coda position.

The last example is the adaptation of [s] to [s]. Consider (26) and (27):

(26) [S] —> [8]

Simon —) [saj.man] §0Phi8 —> léu-fel]
Susan —9 [sugan] Begs —> [Pal-Si]
Sandy —> [sans] Mag —> [maj.khe.§i]
Sandra —> [sagtala] Pierg: —> [phi.ar.§i]
(27) [S] -> [e]

Smith —> [§i.mi.si] Stanley —> [si.t“an.li]/[si.tan.li]
Stone ——> [giton] Scott —> [sikhawthe]
Stanford —) [§i.tan.fu] Lisa —> [li.§a]
Stuart —> [§i.tu.x‘"a] Sarah —-> [gala]

Starr —-> [sitha] Vanesga —> [won.ni.§a]
Steve —> [§i.thi.fu]/[§i.ti.fu]

We can see in the comparison of (26) with (27) that the adaptation [s] —> [s] occurs as

long as [s] is word-initial and followed by another consonant. In addition, the

examples of Lisa, Sarah and Vanessa show that this feature change may occur if the

34

resulting syllable is [sa].

2.2.2 Feature Change in Illicit Consonants

(28) [V] —> [W] 01" [fl

a.
yrcky —> [wej.tchi] Las yegas —> [la.si._wej.tca.si]
yincent —-> [wonsen] Oliyia —) [aw.li.wej.ja]
yiyian ——> [wej.wejan] Kegin —> [khajwan]
Zanessa ——> [wannisa] Dazid —> Ida-Eel]

b.

Steyen —> [si.ti.wen]/[§i.ti.fan] Oliyer —~> [aw.li.fo]

[yon —> [aj._wan]/[aj.fan] Y 20MB —> [Lian]

c.

Steve ——> [sitiru] Exe —> [Lin]

Daye —> [tajfu] Oliye —> [aw.li.fu]

[v] in the onset position is adapted to [w] in most cases. As seen in (28b), when

onset [v] is contained in a syllable such as [van], [var] or [van], it could be adapted to

[f]. In the coda position, [v] is always adapted to [f] with the epenthetic [u] following.

As mentioned in §1.4.], Mandarin obstruents are all voiceless. Therefore, it is

not surprising that English voiced obstruents simply get devoiced in the loanword

adaptation. However, we have found an exception in (28), where onset [v] may be

adapted to [w]. The other exception is [2], which sometimes is adapted to [s] rather

than expected [3].

35

Now consider the adaptation of English voiced obstruents as shown in (29):

(29)

Sorry —> [pa.li] Migouri ——> [mi.§u.li]
Qreg -> [Relej .Re] Louise —> [lu.i.§i]
Adam ——> [jatalj] Chas —) [[sha.§i]
Qerek —> [to.1i.kha] Jazz. —> [tqweei]
Qary —> [Raj .li]

(29) shows that English voiced obstruents all surface as unaspirated voiceless
obstruents. Following Silverman and Yip, I assume that Mandarin speakers, like
Cantonese speakers, are not able to detect the contrast between the two.

Examples of the adaptation of the interdental fricative [0] are given in (30):

(30) [9] —> [8]. [$1 or [91

Keith —-> [tehisi] Matghew —> [majgow]
Kennefl —> [khon.ni._s_i]/[khan.ni.§i] [Rec —> [gi.ow]

A rmur —-> Ua_§9] Timothy -—> [thi.mo.gi]
Heat_h_er —) [xaj.§e] Jonathan —-> [tchaljnaserj]
Cat_herine —> [khaj.§e.lin] Samantha —> [§a.man.§a]

[0] is mostly adapted to [s]. If it is preceded by a front high vowel or glide, [5]

goes through palatalization and becomes [c]. We find that the [0] in Jonathan and

Samantha is adapted to [5] instead of [5].

Now let us turn to the adaptation of [j], [tj] and [d3].

36

(31)L11 —>[9] or 15]

fiow ——) [gow] Bus_h —> [pu.gi]/[pu.§i]
_C_harlotte -—> [calwothe] Sfiaron —> [Qwelwon]/[§a.l“’an]
fleila —> [gila] Sean —> [swo.an]/[ei.an]/[si.on]
Migheue —> [mi. sweat] fire/c —> [§i.rwej.khe]

Cafl —~> [khajgy] Jog/gun -—> [t§“’0.§u.wa]

Nag —> [najpy] cashmere —> [kha.§i.mi.er]

(31) shows that [f] is adapted to [c] or [s]. Careful readers may observe that the
feature [labial] may appear in the output either by adding a secondary articulation or
by appearing in the following epenthetic vowel. To illustrate the latter case, compare
the adaptation of Cash with that of Bush. In the adaptation of Cash, [1] surfaces as [c]
with the rounded front vowel [y] epenthesized after it. In the adaptation of Bush, [1]
surfaces as [c] or [s] with an unrounded epenthetic vowel [i] and [i], respectively.

It is not surprising that [j], [tj] and [d3] are adapted in a similar way. Consider

the examples given in (32) and (33):

(32) [d] -+ [t9] or [Ls]

glip _, [diphu] C_hurc_h_ill —-) [tcfowtcior]
C_has —> [[§_ha.§i] Rag/gel —> [rwejtcfow]
giant —+ [[§_ha.to] Hitgficock —> [ei.te'1y.k“aw.kha]
C_hile -—) [Isili] Lynfl —> [1in.tcfi]/[lin.tcfly]
Richard —> [li.1§_ha]

37

(33) [(13] -> [19;] 0f [LS]

lack —-> [teeltha] Jerry ——> [1,59.li]/[_t_ce.li]
lefi —> [tgefu] Joshua —> [I§w0.5u.wa]
Jazz —> [ewes] Page —> [phej-tei]

(32) and (33) show that [t1] and [d3] are both adapted to [tc] or [L5]. Two things
are worth mentioning here. First, like the adaptation of [f], [labial] sometimes appears
in the output. Second, the Mandarin consonants corresponding to [tj] vary in terms of
aspiration, but those corresponding to [d3] are always unaspirated.

Finally, we see in (34) that [h] is adapted to [x]; if [h] is followed by a front

high vowel, then [x] goes through palatalization and becomes [9].

(34) [h] -> [X] or [91

flilary —> [gi.la.rwej]/[gi.la.li] f_1erb —> [19.p0]
Rilda —> [gotta] flarvey —> ixaweji
flitler —> [cithola] flannah —> [xan-na]
flitchcock —> [Qitghykhawkha] flenry —> LEGO-1i]

2.2.3 Adaptation of Illicit Syllable Structures

While English allows complex onset, Mandarin does not. As a result, when

English forms with complex onset enter Mandarin, this structure has to be fixed.

There are two strategies Mandarin can use — vowel epenthesis and consonant

deletion. However, Mandarin takes the vowel epenthesis strategy in a majority of

C3865.

38

Mandarin does not allow consonant clusters in the coda position either. In fact,

only the two nasals [n] and [13] can serve as coda in Mandarin. Consequently, any

coda consonant other than [n] and [1]] in the English forms has to be fixed. Our data

show that an obstruent coda is always fixed by vowel epenthesis. Consider the

examples given in (35):

(35)
fleven
§£0££
Smith
gee
flank
Slices
Chip
Herb

11111111

[§Qi.won]/[§_i._t_i.fon]
[§i._kfaw.fla1
bio—filial

[kadeikel
[Martial
[arweiklal
[phi-pill

[mpg

C ha_d_
Jefl
Bug/1
Pagg
Casi
Nag/1
Erne§_t
Mas

11111111

[chase]
[tee-LII]
Inn-sfl/[pusi]
[phej-tsi]
[khajgy]
[najsxl
[shim
Imaj.l<"_as_il

It can be seen in (35) that the epenthetic vowels occurring after [5] and [s] are

always [i]; those after [t]/[th] and [k]/[kh] are always [9]; those after [p]/[ph] and [f] are

always round vowels; and those after [9] and [to] are [i] or [y].

If the English form possesses [m] in the coda position, this problematic structure

is fixed by adapting [m] to [n] or [1]] or by epenthesizing the vowel [u] after [m].

Some examples are given in (36) and (37):

39

 

 

   

(36) [m] -> [n] or [1]]

VVIlliam -—) [wej.ljag] modem
Miriam —> [mi.rwej.am Adam
Malcolm —> [majetkhom

(37) [m] —) [mu]

Tim [dim Tom
J im [tcimg] Hume
Sam [swam]

._9
__)

[mo.tjall]
[la-tan]

[tha13.mtfl
[comm]

It is interesting how Mandarin treats the two liquids [r] and [1] if they occur in

the coda position. Our data show that (i) [1] either vocalizes or is adapted to [r] with a

schwa epenthesized before it, and (ii) [r] is either preserved by epnthesizing a schwa

before it or simply deleted. Examples are given below:

(38) [l] vocalizes
Rachel —) [rwej.tchORv_] bagel
Hazel ——> [xaj .rfl]

(39) [fl/[ll -> [9r]

Piegce —> [phiggsi] Nobel
Clage —> [khe.laj.a_r] Miles
Blai; —> [pu.laj .gg] e-mail
Sears -> [ctgsi] Hiflman
Malcolm —-> [majgkhen] Neilson

40

11111

[peikwai

[nwo.pej.o_r]
[majalgsi]
[i.mej.gl_']
[ctgman]

[nigsen]

(40) [r]/[l] —) Q

Ann Al'bor ——) [an.na.paw] 0530a —> [O.san]
Magk ——> [makha] Gogdon —> [ka.talj]
Bagbara ——) [pa.pa.|a] Goldberg —> [kapo]
Stag —) [sitha] Harold —> [xa.lwo.te]
M05! —) [moths]

The occurrence of [r] in the coda position seems to contradict what has been

mentioned in §1.4.3 that only [It] and [1]] can serve as the coda. The issue of coda [r]

in Mandarin will be discussed in §3.5.3. With respect to what causes the

preservation-deletion asymmetry found in the English coda liquids, I will argue in

§3.5.3 that it is attributed to the difference in salience between those preceded by front

vowels and those preceded by back vowels.

41

CHAPTER 3

ANALYSES

3.1 The Nature of Loanword Input

In Silverman (1992), the input to loanword adaptation is viewed as superficial

non-linguistic acoustic signals, and this view results in an analysis in which loanword

adaptation proceeds at two separate, ordered levels: the Perceptual Level and the

Operative Level. Paradis & Lacharité (1997), in contrast, claim that loanwords enter

the borrowing language with structures (i.e., segments and sequences). This view

eliminates the motivation for the existence of the Perceptual Level proposed by

Silverman. Any work on loanwords must deal with the nature of the input. In this

chapter, I will first provide evidence arguing that when English forms enter Mandarin,

the input is best viewed as acoustic signals.

Stating that loanword input consists of acoustic signals entails that it contains

information corresponding to features that are non-contrastive in the lending language.

We therefore wonder if we are able to find those features in the ultimate output. Our

data show that some English non-contrastive features are indeed present in composing

segments of the Mandarin forms.

The existence of English non-contrastive features in the Mandarin forms does

42

 

   

 

not in itself prove the acoustic-signal nature of English loanword input to Mandarin,
because one can claim that the input is a representation at the postlexical level of
English. What we need is to find evidence showing that the input does not involve
structures (e.g., segmental sequences). Our data actually provide such evidence. It is
observed that a sequence of a stop plus a fricative or of a stop plus [r] in the English
form may surface as an affricate, e.g., Reynolds —> [lej.nwo.l_s_i], _7_’I;acy —> [flejci].
We have seen in Chapter 2 that vowel epenthesis is the strategy most frequently used
to repair structures such as complex onsets or illicit codas. If English loanwords enter
Mandarin with their phonological representations (i.e., structures), it is hard to explain
why the vowel epenthesis strategy is abandoned in these cases. In contrast, an
acoustic-signal view to loanword input gives us a straightforward explanation: the
acoustic signals for those sequences are simply interpreted as single segments at the
Perceptual Level.

In §3.l.1 — §3.1.5, I point to those English non-contrastive features that are
found in the Mandarin output. The ‘segment-merging’ phenomenon is investigated in

more detail in §3.1.6.

3.1.1 Aspiration in English Stops and Affricates

Let us begin with English voiceless stops and affricates. English voiceless stops

43

are unspecified for [spread g1] at the lexical level (thus non-contrastive) (Kenstowicz
1994: p.506). At the postlexical level, they are assigned [+spread g1] in the
syllable-initial position but [—spread gl] elsewhere. It is observed in our data that
English syllable-initial voiceless stops in most cases surface as aspirated, e.g., Revin
—> [flajwen], Zed —> [flajta], Reg —> [p_hej.ko], Mifiael --> [maj.k_he]. In addition, in
the appendix of A New English-Chinese Dictionary (1987), where a total Of 238
common English names with form-initial voiceless stops are found, 193 of them
correspond to a Mandarin form that begins with an aspirated voiceless stop, making
up more than eighty percent of the instances (81.1%). This high degree of match
between the syllable-initial position of the English voiceless stops and the aspiration
of their correspondents in the Mandarin forms provides evidence that English
loanword input to Mandarin does contain non-contrastive features.

Our data also show that English syllable-initial voiceless affricates are matched
to Mandarin aspirated affricates, e.g., _C_h_ad -—> [Lsfate], Raghel —) [rwejlglow] (the
only exceptions are Qile ——> [1§i.li] and Churc_h_ill -—9 [tchow.t_gi.ar]). It is certain that
voiceless affricates in English are unspecified for [spread g1] at the lexical level (there
is no contrast between [tj] and [t]"] in English), but whether they are at the postlexical
level assigned [+spread g1] or [—spread g1] in terms of their position is not clear. Based

on my own observation (also personal communication with Dennis Preston) and the

fact that English syllable-initial voiceless affricates are regularly matched to Mandarin
aspirated affricates, I propose that at the postlexical level, English voiceless affricates
are assigned [+spread g1] syllable-initially and [-spread gl] elsewhere. If this is
correct, then the data can be accounted for the same way as English syllable-initial
voiceless stops: [+spread g1] comes into the input and is preserved in the adaptation
process, resulting in aspirated affricates in the output.

Let us now consider English voiced stops/affricates. Like their voiceless
counterparts, English voiced stops/affricates are unspecified for [spread g1] at the
lexical level. At the postlexical level, however, they are assigned [—spread g1] by

default, regardless of where they occur in a syllable. Our data show that English

voiced stops/affricates are matched to Mandarin unaspirated voiceless stops/affricates

(e.g., _G_ary —> [Rajli], Ted —> [thajla], .lefl—> [lgefu], Page —> [pheitgin This once

again shows what is assigned at the postlexical level of English is found in Mandarin
output.

English voiceless stops/affricates in the coda position are unspecified at the

lexical level and assigned [—spread g1] at the postlexical level. We thus expect their

Mandarin correspondents to be unaspirated. Surprisingly, our data show that English

voiceless stops/affricates in this particular position are matched to Mandarin aspirated

stops/affricates (e.g., Hitficog/g —) [ci.t_9:y.khaw.k_ho], Mag —> [maj._t29]). I suggest this

45

 

is attributed to the fact that English voiceless stops/affricates in the coda position are

released (personal communication with Eric Zee). Though the release of air is much

weaker and shorter than the aspiration of syllable-initial voiceless stops/affricates,

Mandarin speakers still perceive it as aspiration. In fact, there are some cases in our

data in which English voiceless stops in the coda position are deleted. For example,

Raymond surfaces as [1ej.melj] and cos; as [khasi]. In most of these cases, the deleted

stops are in a consonant cluster. I assume that these deleted stops are unreleased and

therefore not perceived by Mandarin speakers at all. The result is that they are not
included in the input to the loanword adaptation.

Note that whether we assume [spread g1] to be a binary or a monovalent feature

does not affect our analysis. In a system where [spread g1] is viewed as monovalent,

English voiced stops/affricates are assigned ‘nothing’ instead of [—spread g1] at the

postlexical level. Their Mandarin corresponds are unaspirated simply because

‘nothing’ is in the input.

3.1.2 Lip Rounding in English [1']
The English fricative [j] is strongly labialized (Ladefoged 2001: p.53). Our data

Show that the feature [labial] sometimes appears in a Mandarin syllable or segment

that is matched to [1]. Some examples are given in (41):

46

 

 

 

(41)

Input Output

can —> [khajgy]
Nag/1 —9 [naj.cy]
Sharon —> [gfelwen]
Shirley ——> [deli]
Mic_helle —> [micfeer]

(41) shows that [l] is matched to [cy] or [cw]. In both cases, the feature [labial] is
observed. Since [labial] does not contrast [1] from other segments, it is assigned to [j]
at the postlexical level, probably by a language-specific rule. The presence of [labial]

in the Mandarin syllables/segments indicates that the English input contains this

feature as well.

3.1.3 Syllabicity in English [1]
The next English non-contrastive feature to consider is syllabicity in [1]. There is
an English allophonic rule which says that [l] is syllabic (thus [+syllabic]) at the end
of a word when immediately preceded by a consonant (e.g., paddle [phaedl], whistle
[wrsl 1, kennel [khenlD (Ladefoged 2001: p.58). Our data show that English [l]s in this

particular context are matched to [ow] or [0], but those occurring elsewhere are not.

Some examples are given below:

47

 

‘1

 

 

 

 

 

 

(42)

Input Output
a. Rachel —> [rwej.tcho_w_]
Hazel —> [xaj.ro_vfl
bagel —> [pej.kwg]
b. Lisa —9 [lisa]
Clare —> [khalajer]
c. Miles —> [majggsi]
Nobel ——> [nwo.pej.e_r]
d. Goldberg —> [ke.po]
Albright —> [ow.pu.laj.tha]

In (42a), the English [l]s, which occur at the end of a word and immediately
follow a consonant, are adapted to [ow] or [o]. The examples in (42b) — (42d) show
that English [l]s which occur in other contexts either surface intact (42b), are adapted
to [or] (42c), or get deleted (42d). Following Yip (1993: p.273), who observes a
similar phenomenon in English loanwords in Cantonese, [ow] and [o] in (42a) are not
epenthetic; instead, they are realized by vocalization of the English [l]s4. The reason is
simple: epenthetic vowels triggered by unsyllabified [tch], [r] and [k] are never found
to be [ow] or [o] in Mandarin loanword adaptation. Given that [+syllabic] in English
[l]s is assigned at the postlexical level, its presence in the Mandarin output (contained
in the vowels) argues for the involvement of English non-contrastive features in the

input

 

4 Alan Munn (personal communication) points out that coda [l] vocalization is common in some
varieties of English. especially in closed syllables such as [wolf] ‘wolf’. It is not unlikely that the
English input is actually a vowel rather than [1] in certain words.

48

3.1.4 Velarization of English Coda [1]

It is known that English [1] is velarized when it occurs in the coda position5 (e.g.,
file [fajl], field [fild]) and that the relevant feature, [dorsal], is assigned by a
context-sensitive rule at the postlexical level. Our data show that if English coda [l] is
preceded by a back vowel, the [l] is deleted in the adaptation; whereas, if it is a front
vowel that precedes the coda [1], then the [l] is preserved and adapted to [or]. Consider

the following examples:

(43)
Input Output

a. Goldberg —> [ka.po]
Harold —> [xa.lwo.te]
Albright —> [ow.pu.laj .tha]

b. Miles —> [maj.gsi]
Neilson —> [nigsen]
Nobel —> [nwo.pej.e_r]
Hiflman —> [ci.9_r.man]
Malcolm ——> [maj .9_r.khan]

In (43a), the [Us in the English forms are preceded by a back vowel and have no

correspondents in the Mandarin forms. (43b) shows that if the preceding vowel is a

front one, [1] is retained instead of being deleted. It has been widely stated in the

 

5 Alan Munn (personal communication) claims that English [1] preceded by a front vowel is not
velarized or is only slightly velarized. According to Ladefoged (2001: 84), however, a front vowel in a

word such as [hi1] ‘heel’ or [heel] is retracted. This suggests that English coda [l] is velarized regardless
of the quality of the preceding vowel.

49

literature that liquids are unsalient and thus sometimes not parsed. For example, Fay
and Culter (1977) claim that liquids possess vowel-like formants so that they fail to
saliently stand out from the surrounding vowels. Now assume that the English [l]s in
(43a) come into the input with the feature [dorsal]. Since the preceding back vowels
possess the feature [dorsal] as well, it follows naturally that it is even harder for those
[l]s to saliently stand out (they share more properties with the preceding vowels).
When the adaptation process arrives at the Operative Level, where the Mandarin
phonotactic constraints hold, Mandarin chooses to delete them instead of preserving
them. In contrast, the [HS in (43b) are more salient than those in (43a) for two reasons:
(i) the preceding vowel is not a dorsal but a coronal, and (ii) a schwa might be
inserted between the [l] and the preceding vowel. As a result, the [Us in (43b) are
more likely to be preserved through the adaptation. The examples in (43) indicate that
whether English coda [l] is preserved or deleted in the loanword adaptation is related
to the quality of the preceding vowel. That is, a velarized [l] is deleted only if the
preceding vowel contains the feature [dorsal]; otherwise, it is preserved. Since
English [1] obtains [dorsal] at the postlexical level, this close relation argues for the

claim that the loanword input contains English non—contrastive features.

50

 

 

 

3.1.5 Nasalization of English Vowels Before N asals

In English, vowels are nasalized if followed by a nasal consonant (e.g., cgn
[khén], slng [519]). Since the feature [nasal] is predictable in such vowels, it is
regarded as non-contrastive. Our data show that this particular feature may appear in

the output. First of all, consider the examples given in (44):

(44)

Input Output Input Output

Ben —> [pan]/[pan.an] King —) [tcin]/[tcin.an]
Dan —> [tan]/[tan.en] Lynn —> [lin]/[linen]
Ken —> [khan]/[khan.an]

We see in (44) that the English forms are all monosyllabic and each has two

Mandarin equivalents, one monosyllabic and the other bisyllabic. I will argue in

§3.5.4.2 that if the English form is monosyllabic, it may be augmented by vowel

epenthesis and surfaces as bisyllabic. Assume this is correct for the moment. Now

compare the two Mandarin forms for each English input in (44). It is certain that the

epenthetic vowel is a schwa. A question then arises: Is the syllable [on] as a whole

added at the end of the monosyllabic form, or is a schwa inserted between the vowel

and the coda [n] with another [n] appearing right after the vowel? There is evidence

supporting the latter suggestion. Consider the following examples:

51

(45)

Input Output Input Output
Main —> [mejon] Jean -> [tchin]/[tci.en]
Wayne —-> [wej.on] Boone —> [pu.en]

All the English forms in (45) contain a tense vowel and are augmented in the

loanword adaptation. We can clearly see that the augmentation is realized by

epenthesis of a schwa and the location of the epenthesis is between the vowel and the

coda [n]. An assumption where the syllable [on] is added at the end of the forrn has to

explain the deletion of the [n], but [It] deletion (or more generally, deletion of nasal

segments) is never found in English loanwords in Mandarin.

Having identified the location where the schwa is epenthesized, the question now

is the status of the [n] that occupies the coda position of the first syllable. Put

differently, where does it come from? An epenthetic status of the [n] is not supported.

Take the adaptation Lynn —9 [linen] as an example. We all know that loanwords are

adapted minimally. Given that [1i] is a well-formed Mandarin syllable, epenthesizing

an [n] at the end of the syllable does not seem to be motivated (i.e., Why does Lynn

not surface as [li.en]?).

If we adopt the assumption that English loanword input to Mandarin contains

non-contrastive features, this question has an answer: the observed [n] is derived from

the feature [nasal] involved in the English vowel. I argue that this ‘unpacking’

52

process6 is triggered by the two-mora requirement in Mandarin (see §1.4.3). We
know that in English a tense vowel/diphthong possesses two moras while a lax vowel
possesses one. When English forms such as those in (44) enter Mandarin and undergo
augmentation by schwa epenthesis, the first syllable is judged as ill-formed by
Mandarin because its rhyme only possesses one mora. To fix this ill-formed syllable,
the most economical way seems to unpack the nasalized vowel, letting the feature
[nasal] involved in the vowel surface as [II]. By doing so, the resulting syllable has
two moras and meets this particular requirement imposed by Mandarin.

As for the examples in (45), I suggest that the feature [nasal] is deleted during
the adaptation. When English forms such as those in (45) enter Mandarin and are
augmented to bisyllabicity by schwa epenthesis, the first syllable possesses two moras
and satisfies the two-mora requirement. Since Mandarin does not allow nasalized
vowels and no extra slot is available for the feature [nasal] to surface as [n], it has to

undergo deletion.

3.1.6 Segment Merger
I have shown in §3.l.1 — §3.l.5 that English loanword input to Mandarin

contains features that are non-contrastive in English. In this subsection, I will provide

 

6 This term is given by Paradis & Prunet (2000), who argue that ‘contrastiveness in the source
language determines the presence or absence of unpacking in the borrowing language.’ The adaptations
in (44) seem to be counterexamples to their claim.

additional evidence to argue that the input enters Mandarin without structures (i.e., the

input consists of acoustic signals).

Compare the underlined segments in the English forms with their Mandarin

correspondents:

(46)

a. b.

Input Output Input Output

Reynolgfi —> [lej .nwo.t_si] Racy —> [fled-Gil

schng -—) [pu.lsi] Patrick —> [phajtsl‘y-khai

Roberts —> [lwo.po.l§i] Sanchez -+ lean-e301

13er _;. [P611921] Alexanfla —) [ja.li.§an.1§:0]
Sina_t_r_a —) [cin.na.lc:y]
Sl_r_eep —> [§i.[§h_wej .phu]
fluman —> [filamen]
St_reisand —> [siflejsan]

(46a) shows that a stop and an immediately following fricative in the English
forms are matched to a Mandarin affricate. We observe a similar situation in (46b): a
stop and an immediately following [r] correspond to a single Mandarin affricate.
There is no doubt that in (46a) the two English segments merge to [ts]/[tch] in the
adaptation. For (46b), however, it is not that obvious. I would say that they merge as
well, since if the liquid [r] were deleted, the preceding stop would surface as [t] rather

than as an affricate. Paradis & Lacharité’s view that loanwords enter the borrowing

54

language with structures fails to account for this fact. By their view, the input is
immediately interpreted as a phonological representation by the borrowing language
(p.380). Since the input contains segments that are in a sequence, we would expect a
vowel (a schwa) to be epenthesized between the stop and the fricative/[r] in (46).
However, this is not what we have seen. (47) illustrates their view using the

J

adaptation Of Roberts and Tracy:

(47)
Input Loanword adaptation7 Output

(with structures)

[ra.bert_s] Roberts —> Epenthesize [e] —> *[lwo.po.§3._si]
between [t] and [s]

[th_rej .si] Tracy —> Epenthesize [a] —> *[therwejci]
between [th] and [r]

An acoustic-signal view to the nature of loanword input is able to account for the
merger of the two English segments in the adaptation. In this view, the acoustic
signals for the stop and the following fricative/[r] are represented as an affricate at the
Perceptual Level. At the Operative Level, a vowel is epenthesized if the affricate is

assigned a syllable node. This view is exemplified in (48):

 

7 Only the schwa epenthesis in question is shown here. Other types of epentheses and processes of
segmental adaptation are ignored.

55

(48)

Input The Perceptual The Operative Output
(acoustic Level Level
signals)
Roberts —> [ra.per.ts] —> Epenthesize [i] —> [lwo.po.t_si]
after [ts]
Tracy —> [tshwejsi] —> N/A —-> [tshwejci]
3.1.7 Summary

We have seen that English non-contrastive features such as [+spread g1] in
stops/affricates, [labial] in Ll], [+syllabic] or [dorsal] in [l] and [nasal] in vowels may
appear in the Mandarin output. We have also seen that two English segments of
certain types may merge to an affricate in the adaptation. These facts are in
conformity with an acoustic-signal view to the nature of loanword input. Paradis &
Lacharité’s view fails to account for the segment merging phenomenon shown in the

last subsection, since it assumes that loanwords enter the borrowing language with

structures, including segments and sequences.

3.2 At the Perceptual Level

3.2.1 Licit Consonants

I have argued above that English loanword input to Mandarin is best viewed as

acoustic signals. Following Silverman (1992), the acoustic signals are divided into

56

segment-sized chunks for which the closest feature matrices from the Mandarin
inventory are provided. Our data show that if the English consonants exist in
Mandarin, they usually surface intact. For example, most English aspirated voiceless
stops surface without any change. Nevertheless, some exceptions are found. For
example, although both English and Mandarin have the labial nasal [m], it is mapped
to [n] or [1)] when it is assigned to the coda position. A few examples show that a
vowel is epenthesized to the right of it (e.g., Tom —> [that].r_n_g). These examples will
be discussed in §3.3. Clearly, coda [m] surfaces differently because Mandarin only
allows [n] and [13] in the coda position. Following Silverman (p.297), I assume that
the acoustic signal of every English [m] is represented as [m] at the Perceptual Level;
when the loanword adaptation process arrives at the Operative Level, where the
Mandarin phonotactic constraints hold, those assigned to the coda position undergo
feature change and turn into [n] or [1]]. (49) shows that no matter in what position the

English [m] occurs, its acoustic signal is always represented as [m] at the Perceptual

Level.

57

(49)

Input The Perceptual Level8 The Operative Level
a.

Smith —> [s._n_ri.s] —> [§i._lgi.si]

Mary —) [men] —-) [mali]

b.

Adam —> [e.ta_m_] —> [ja.tag]

VVzlliam ——> [wi.liom] -—> [wej.l’ag]

In (49a), the acoustic signal of the input [m] is represented as [m] and assigned to
the onset position; in (49b), the acoustic signal of the input [m] is also represented as
[m] although it is assigned to the coda position.

The phonemes [r], [s] and [n], which also exist in Mandarin, sometimes undergo
feature change as well. Our data show that [r] and [s] in onset position and [n] in coda
position sometimes surface as [l], [s] and [1]], respectively. For example, Magi —-)
[ma.li], Lisa —> [11.sa] and Damog —> [taj.mag]. It is puzzling that the adaptations take
place because in Mandarin [r] and [s] are allowed in the onset position and [n] is
allowed in the coda position. Just like English [m], I assume that at the Perceptual
Level, the acoustic signals of English [r], [s] and [n] are represented as such
irrespective of their position, and it is during the Operative Level that they may
undergo feature change. Nevertheless, [r], [s] and [n] are adapted for different reasons.

I suggest that the following two factors trigger the adaptation: (i) the choice of

 

8 According to Silverman (1992), post-vocalic [s] is assigned a syllable node at the Perceptual Level
(see §1.2.l).

58

 

Chinese characters; and (ii) some syllables simply cannot surface in Mandarin
because they lack corresponding Chinese characters. These two factors are also able
to account for many exceptions, including those in which English aspirated

stops/affricates surface as unaspirated. In §3.4, I will show evidence to argue for my

claim.

(50)

Input The Perceptual Level The Operative Level
a.

Redford —> [flejfor] —> [flejfu]
S imofl —> [§aj.mag] —> [saj.meg]
Zed —> [flet] —) [Lhajta]
b.

May ——> [magi] —> [ma.li]
Liga —> [li.§a] —> [li.§a]
Damon —~> [tej.mag] —> [taj.meg]
Iony -—> [tho.ni] —-) [lonni]

3.2.2 Illicit Consonants

Our data also show that if the English consonants do not exist in Mandarin, they
are mapped to the closest Mandarin segments. For example, although English
obstruents contrast in voicing, Mandarin obstruents do not. Following Silverman
(1992) and Yip (1993), I assume that Mandarin speakers do not even detect the

contrast. Therefore, the acoustic signals of English voiced obstruents are represented

59

 

as voiceless obstruents at the Perceptual Level. Two things are noted here. First,
English voiced stops and affricates all surface as unaspirated in spite of the fact that
Mandarin voiceless stops and affricates contrast in aspiration, as we have seen in
§3. 1 . 1. Second, the voiced labio-dental fricative [v] is mapped to [w] or [f].

The interdental fricative [0] and the glottal fricative [h] are mapped to [s] and [x],
respectively. If they are followed by the high front vowel [i], they surface as [9].
Similar to [s], [0] sometimes surfaces as [s] due to selection of the character.

Let us now consider [j], [tj] and [d3]. Since [1] differs from [tj] in
presence/absence of the [t] and [tj] differs from [d3] in voicing, only [j] is discussed

here. Our data show that [1] surfaces as [c] or [s]. However, the feature [labial]
sometimes is observed in the output, e.g., Mic_h_elle —> [migfieer]. Based on the fact
that English [j] has the non-contrastive feature [labial], I assume that the acoustic
signal of [j] is represented as [gw] or [pw]. This raises a question: Why is the feature
[labial] deleted in some cases (e.g., Sfieila —-> [gila], Shrek —> [§i.rwej.kha])? Again,

like onset [s] and [r] may surface as [s] and [l] or coda [It] may surface as [1]], I

suggest that it is character selection that triggers the deletion.

60

 

(51)

a. Voiced obstruents

11111

1

1111

Input The Perceptual Level
flurry —) [peri]
Adam —> [e.tem]
Qary —-> [Red]
M igsouri —> [masuri]
Stezen —> [sti.fen]/[sti.wan]
b. [9]
Arflur -—> [anger]
c. [h]
flerb —> [xerp]
d. [1]. [t1]. [d3]
flaron ——> [fle.ran]/[§:e.ran]
C_hip —> [fliphllltfliphl
caas —> west/wen
lerry —> Morn/[lien]
3.3 Mandarin Palatals

As we have seen in §1.4.], the palatals [t9], [tch] and [9] are allophones in
Mandarin. In many studies, they surface by the phonemes assimilating to a following

high front vowel or glide. Duanmu (2000: p.28, 81), however, claims that the

phonemes and a following glide merge into a single segment. (52) and (53) show the

difference using the dental sibilants [ts tsh s]:

61

The Operative Level
[pa.li]

[iataul

Ital-1i]

[mi.§u.li]

[§i.ti.len]/[§i.ti.wan]

[rape]

[fle.lwen]/[§g.lwen]
[195-1931]

[Isla-si]
[lge.1i]/[1§a.li]

 

 

 

(5 2) Others

Underlying Representation Surface Representation
Its tsh s/ + /j/ —> [tc tr]:h c] + [1]
Its tsh s/ + /q/ —> [to t9h Q] + [II]

(53) Duanmu

Underlying Representation Surface Representation
/ts tsh s/ + /j/ —) [t9 to“ 9]
Its tsh s/ + /q/ -> [WW I?“ 9w]

In (52), the palatals and the following glide are independent segments in the
surface representation, but in (53), the phonemes and the following glide merge into a
single segment. In this section, I will draw evidence from our data to argue for
Duanmu’s view. First of all, consider the adaptations in (54) where the output forms

are transcribed in both ways (For the sake of convenience, I discuss [cw] only).

(54)
Input Output

Others Duanmu
mow —> [210W] [cowl
Shirley —> [93e.li]/[§je.li] [fle.li]/[§e.li]
Qiarlotte —> [gialwothe] [pa.lwothe]
Mighelle —-) [miggear] [migfieer]
Simon —-) [ggelwen] [flelwen]

In (54), English [1] is mapped to [c]/[cw] in Duanmu’s transcription but to

 

[ej]/[cu] in other people’s. As suggested by Silverman (1992), the acoustic signals are
parsed into segment-sized chunks for which the closest native segments are provided.
lntuitively, the acoustic signal of [1] should be parsed into one chunk rather than two,
and therefore only one segment from the borrowing language can be provided. I have
assumed that when English [j] enters Mandarin, [cw] is one of the segments that fill
the chunk. Duanmu’s view has no problem with this assumption: The acoustic signal
of [l] is represented as [9"] at the Perceptual Level, and then it surfaces with no
change or as [9] due to selection of the character. According to the views of others,
however, the occurrence of the glide [j]/[q] in the output cannot be explained.
Assuming an epenthetic status of [j]/[t{] also fails, in the sense that loanword

adaptation is minimal and therefore there is no reason to epenthesize them.

(55)9
Input The Perceptual Output

Level
show —-> [9:0] -—) [gow] *[gjow]
S_lz_irley —> [glenli] —> [gem/[cell] *[gge.li]/[§je.li]
ghanmte —> mallet") —> [calwotha] *[gjalwotho]
Mic_h_elle —> [mig’lel] —> [micjeor] *[mi.9ge.or]
Simon —> [fleran] —> [flelwan] *[Qgelwen]

Let us now consider English [j] that is assigned to the coda position. Some

 

9 The processes at the Operative Level are omitted in this chart.

63

 

examples are given in (56):

(56)
Input Output

Others Duanmu
Bus/.1. —> [rm-91] [DU-91]
Cash —> [khajgy] [khajgy]
Nag —-) [naj.§y] [najgy]

Although Duanmu and other people have the same transcription for the output
forms, only Duanmu’s view can explain the realization of [9y]. As mentioned above,
the acoustic signal of [j] is represented by [cw] at the Perceptual Level. When the
loanword adaptation arrives at the Operative Level, a nucleus is epenthesized after [9“]
because [cw] has been assigned to the coda position and Mandarin does not allow this
structure. The [cw]+nucleus sequence then surfaces as [9y] through assimilation. Note
that the epenthetic nucleus may be realized as [i], as in Bufi —> [pu.gl], which I
assume is due to selection of the character.

In theories other than Duanmu’s, [9"] cannot be used to fill the chunk because it
is not a legal segment. The segment that is chosen for the chunk might be [c], [s] or
[p]. At the Operative Level, a nucleus is epenthesized since none of them is a legal
coda consonant. A problem now emerges: the epenthetic nucleus is realized as [i] after

[9], as a dental apical vowel after [5] and as a retroflex apical vowel after [s]. The

. '.'_{

 

consequence is that the attested output [cy] would never surface.

(57) (‘N’ denotes an epenthetic nucleus)

Input

The Perceptual
Level

The Operative
Level

Output

Input

The Perceptual

Level

The Operative
Level

Output

Cash

h
k e.s

khajsN

h . .
*k aysr

Nash

“6.8

naj .sN

*najsi

Others

Nash
1

ne.§

naj.§N

*naj.§i

65

Cash

khec

khach

khajci

Nash

nee

naj.cN

*naj.ci

Duanmu

Input Cash Nash
1 1
The Perceptual khet;w ne.cw
Level
1 1
The Operative khajch najch
Level
1 1
Output khajcy naj .cy

In order for the syllable [cy] to surface, one has to assume that the acoustic
signal of English U] is represented as [9”] at the Perceptual Level. Since [9“] is the
outcome of the phoneme [s], [s] or [x] merging with the glide [q], Duanmu’s proposal

is supported.

(58)

/s s x/+/q/
1

[j] —> Acoustic Signal —> [cw]

I will also argue in this thesis that the loanword adaptation not only operates on
the output of the postlexical level of English, but also operates on the output of the

postlexical level of Mandarin. I will show that the acoustic signals are interpreted as

66

 

phonological representations using Mandarin allophones, rather than its phonemes.
The above analysis implies that the loanword adaptation operates on the output
of the Mandarin postlexical level (that is, allophones), which is in contradiction with
Paradis & Lacharité’s view that loanword input enters into the borrowing language’s
lexicon. I agree with Yip (1993), who argues that loanword adaptation does not
constitute a separate grammar from the native phonology. As I will show in §3.5, the
Operative Level of the loanword adaptation has the same function as the lexical level
of the native phonology. At both levels, the Mandarin phonotactic constraints exert the
influence. However, the above analysis suggests that the segments that are used to
make up the inputs of the two levels are different. For the Operative Level of the
loanword adaptation, the input is the output of the Perceptual Level, which is made up
of Mandarin allophones that most closely approximate the acoustic signals. For the
postlexical level of the native phonology, the input is made up of the phonemes. The

following diagram illustrates the relation of the loanword adaptation to the native

phonology:

67

 

(59)

 

Mandarin Inventory

1

Lexical Level

1

Postlexical Level

1

Acoustic Signals ___> Perceptual Level

1

Operative Level

1

Output

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

3.4 Character Selection

We have seen that English non-contrastive features may appear in the output and
that this fact can be explained only if the input to the loanword adaptation is assumed
to contain these features. Now the question is how to explain the examples where
these features do not show up in the output. In this section, I will argue that these

features may be deleted during the adaptation. To account for these ‘exceptions’,

consider the examples given in (60) and (61):

68

 

(60)

 

 

Input Output

a. flaula —> [pola]
Iony —> Ltonni] c.f. [fianni]
Kennedy —-> [Ran.naj.ti]

b. Qile —> [Lsili]
Churgjlill —> [tchowlgier]

c. Sleve —> [sifiiru] c.f.[§i.li.fu]
Slanley —> [siflanli] c.f. [si.lan.li]
Sgott —-> [§i.k_haw.tha]

(61)

Input Output

fleila —9 [gi.la]

Sharon —> [salwan] c.f. [ctelwen]
Shrek —9 [§i.rwej.kha]

Sean —> [gi.en]/[§i.an] c.f. [§_"fo.en]
Bush —> [pu.gi]/[pu.§i]

Cashmere —> [kha.§i.mi.ar]

(60a) and (60b) show that English syllable-initial voiceless stops and affricates

are de-aspirated. In (60c), English voiceless stops which are preceded by [s] are
matched to Mandarin aspirated stops. (61) shows that English [j], which has been
assumed to contain the non-contrastive feature [labial], surfaces as [c] or [s]. In both
cases, [labial] are absent. To account for these exceptions, I propose that selection of

Chinese characters is involved in the loanword adaptation. In the Chinese language, a

character corresponds to a syllable. In addition, each character possesses a certain

69

meaning. It makes sense that when an English name enters Mandarin, Mandarin
speakers prefer a Mandarin equivalent that is composed of proper characters even if
the equivalent is not so close to the input. Take [any —-) [logni] as an example.
Although the word-initial voiceless stop is de-aspirated, when assigned the first tone,
[tOtJ] may correspond to the character ‘E'E’, which means ‘east’ and is quite common
for male names in China. [thaljfll] is another Mandarin equivalent to Tony. In this case,
the feature [+spread g1] in the word-initial voiceless stop is kept. However, we can see
that instead of having the closest vowel [o], the first syllable has the vowel [a]. The
reason that [o] is replaced with [a] is because when assigned the first tone, [than] may
correspond to the character ‘YZJ’, a Chinese surname (In a Chinese name, the surname
goes before the given name). The choice between [torj.ni] or [tha13.ni] as Tony’s
Mandarin equivalent is really personal preference, but no matter which one is chosen,
it reveals that the composing characters are taken into account.

Let us consider another example. I have assumed that the acoustic signal of
English [1] is represented as [cw] or [§w] at the Perceptual Level. Our data show that
Sharon surfaces as [g’lelwan] or [salwan]. The fact that the feature [labial] is absent in
the latter case may be attributed to consideration of Chinese characters as well. For
the syllable [swa], the most common character has the meaning ‘to brush’ when

assigned the first tone (‘hill’) and ‘to play with’ when assigned the third tone (‘EU

70

(there are no characters for this syllable with the second or fourth tone), and neither is

appropriate for names. In contrast, when assigned the first tone, [5a] may correspond

to the character ‘71}, which is commonly used for female names. Once again, this

example shows the involvement of character selection in the loanword adaptation.

Consideration of Chinese characters also accounts for cases in which licit

consonants undergo feature change. Consider the following examples in which [n] is

adapted to [n] and [s] is adapted to [s]:

(62)

Input Output Input Output

a. b.

Raymond —> [lej .man] Stone —> [§i.tor)]

Damon —> [taj .mog] Stanford —> [§i.tan.fu]

Jonathan ——> [tchannasan] Stuart —> [§i.tu.xwa]

Vincent —> [wanseg] Sam —> [sanmu]
Vanessa —) [wan.ni.§a]
Lina —> [li.§a]

At first glance, the adaptations are puzzling because Mandarin has both [It] and
[s] and allows [n] in the coda position and [s] in the onset position. As suggested
above, it is consideration of Chinese characters that triggers the adaptations. For
example, in Stanford —> [§i.tan.fu], the word-initial [5] surfaces as [5] instead of [8]
simply because, when assigned the third tone, [si] may correspond to the character

‘E’, a Chinese surname. This ‘sumame’ reason explains many other cases in which

71

word-initial [s] followed by another consonant surfaces as [s]. In addition, masculinity
or femininity of the characters sometimes becomes an influence as well. For example,
Raymond surfaces as [lej.men] rather than [lej.men] and Lina surfaces as [li.§a] rather
than [li.§a] because the characters corresponding to [men] (2%”) and [pa] (“T/F) look
masculine and feminine, respectively.

In addition, a licit consonant may get adapted because the syllable where it
occurs simply cannot surface.1 0 Consider the following examples where English

onset [r] surfaces as [l] or [r“].

(63)

Input Output Input Output

a. c.

Rachel —) [lej.tchow]/[_r:ej.tchow] Richard —) [li.t§ha]

Redford —) [flejfu] Ricky —> [litchiI/[ficjtchi]

Shgek —) [§i.r_":ej .kha] Rita —> [li.tha]/[_r:ej .tha]
Hilany —> [ci.1a.li]/[ci.la.fl ej]

b. Jeffgey —> [tcej.fo.li]/[tcej.fo.fl_ej]

Robin —) [la.pin] Enic —> [aj.li.lt"a]/[aj.t_“’_ej.lt"a]

Rice —) [laj .si]

Brian —-> [pu.laj.an]

The phoneme [r] exists in Mandarin and is allowed in the onset position.

However, (63) shows that the input [r], which is assigned to the onset position at the

 

’0 According to Duanmu 2000:§3, the majority of expected Mandarin syllables are missing. Some

missing forms are systematic gaps. For example, a Mandarin syllable cannot have two segments that
are. both [+round]. Some missing forms are accidental gaps. For example. while [m’e] ‘wipe out’ and
[ph’e] ‘skim’ are well-formed Mandarin syllables, *[f‘e] is is not. Thus, *[f’e] is probably an accidental

gap-

72

"—3

 

Perceptual Level, surfaces as [l] or [rw]. Consider the English syllables that contain [r].
We find that the [r] is followed by [ej], [8] (63a), [a], [aj] (63b) and [I] (63c). If we
further assume that the acoustic signals of these vowels most closely approximate
Mandarin [ej], [e], [a], [aj] and [i], respectively, then at the Perceptual Level, the
syllables in question are represented as [rej], [re], [ra], [raj] and [ri]. Because these
syllables cannot surface, the easiest way for Mandarin to fix them seems to be to turn
the onset [r] into [1], meaning that only the value for [lateral] changes. In many
examples in (63), especially in (63c), the English [r] surfaces as [rw]. I assume that
this is due to consideration of characters. Careful readers may have found that in (63c),
if the English [r] surfaces as [rw], the following vowel is always [ej]. Given that
English [ri]/[rl] surfacing as [twej] instead of [1i] violates the ‘minimal change’
principle, only selection of character is able to account for it.

Some scholars claim that English onset [r] involves lip rounding (e.g., Duanmu
(2000: p.26) and many others). If this is true, then the syllables in question should be
represented as [rwej], [rwe] (63a), [rwa], [rwaj] (63b) and [rwi] (63c) at the Perceptual
Level. Note that, according to Duanmu’s proposal, [rw] is a legal Mandarin segment,
which is derived from the merger of [r] with a following glide [w]. However, except
for [rwej], none of those syllables has a corresponding Chinese character. As shown in

(63), only the first syllable of Rachel surfaces intact (fichel —-) [fiej.tchow]); other

73

 

syllables all surface with featural change made either to [rw] or to the vowel. This
supports my claim that a licit consonant gets adapted because the syllable where it
occurs corresponds to no character and cannot surface.

To sum up, I have argued that character selection is able to account for some of
the examples that are considered exceptions, and the reason is that Mandarin speakers
would like an output form that is composed of proper characters even if they deviate
from the input. In addition, some syllables simply cannot surface in Mandarin because
they lack a corresponding character. The consequence is that either the consonant or
the vowel has to be adapted. It should be noted here that although the output may
deviate from the input due to character selection, there seems to be a limit in terms of
how different they can be. The main principle in loanword adaptation is to keep the
output as close to the input as possible. When it comes to Mandarin loanword
adaptation, this principle is not followed strictly because of the influence of character
selection. However, as I have shown, this principle is not totally ovenidden; otherwise
we would have seen adaptations where the input and the output are entirely different.
Our data show that only a few types of feature changes are made when the character

selection factor exerts its influence. They are listed in (64):

74

 

(64)

Type Example
a. [+spread g1] <—> [-Spread g1] [any —> [lorjni], Sleve —> [stifliru]
b. [labial] as a secondary articulation H Sleila —-) [gila], Hilagy —> [cilaflej]
O
c. Coronal —) Dorsal (nasals) Raymond —> [lej.man]
d. [—retroflex] —~> [+retroflex] Lina —> [li.§a]
e. [—lateral] —-) [+lateral] Robin —> [la.pin]

A question arises: Why are only these types of feature changes tolerated? Given

that loanword adaptation is minimal, it seems like some types of feature changes

make more difference in perception than others. For example, we never find feature

changes such as [+continuant] <—> [—continuant] occurring when character selection

comes into force. This issue is left open here and it deserves further research in the

future.

3.5 At the Operative Level

After the acoustic signals turn into native segments at the Perceptual Level, the
loanword adaptation process arrives at the Operative Level, where true phonological
operations apply in order for the segment strings produced at the Perceptual Level to

surface in conformity with the native phonotactic constraints. In this section, I will

75

 

 

discuss those phonological operations and give an analysis within the framework of

Optimality Theory.

3.5.1 Mandarin Syllable Structure Constraints

Syllable structure constraints are a set of constraints about the well-formedness

of syllables in a given language. For Mandarin, they include the following four

constraints:

a. *COMPLEX: Complex onset or coda is not allowed.

b. N UCLEUS: Syllables always have nuclei. Nuclei are always vocalic.

c. CODA CONDITION: In the coda position, only [n] or [n] is allowed.

d. MINSYLL: A full syllable has two moras.

These constraints are undominated, since a well-formed Mandarin syllable has to

satisfy all of them. In addition, these four constraints cannot be ranked with each

other.

3.5.2 Vowel Epenthesis

As shown in §2.2.3, when English forrns enter Mandarin, consonant clusters

76

pan

 

Ir

assigned to the onset position and any obstruents assigned to the coda position are

fixed by means of vowel epenthesis. Some examples are given in (65):

(65)

Input The Perceptual The Operative Level
Level

Eric —> [e.rikh] —> [aj.li.kho]/[aj.rwej.kha]

Glen —-> [klen] —~) [ke.lan]

Ernest —) [er.ni.st] —-) [e.ni.si.tha]

Frank ——> [frenkh] —> [fu.lan.khe]

Since consonant deletion is not used as a strategy to fix these ill-formed
structures, MAX-C-IO outranks DEP-V—IO. The tableau in (68) shows that for the
inputll [klen], which possesses two consonants in the onset position, the candidate

[kelan] is selected as optimal:

(66) MAX-C-IO: Input consonants must have output correspondents.

(67) DEP-V—IO: Output vowels must have input correspondents (No epenthesis of

vowels).

 

” It means the input to the Operative Level.

77

 

(68)

 

 

 

 

 

 

 

 

[klan] ‘Glen’ *COMPLEX CODA CONDITION MAX-C-IO DEP-V—IO
6' [kalan] *(g)
_ [klan] *(kl)!
[lan] *(k)!
[kan] *(1)!

 

 

 

 

Lin (1992) observes the following three types of epenthetic vowels in Mandarin

loanword adaptation: the apical vowel [i], the rounded vowel [u] and the schwa. Some

examples found in our data are given in (69):

(69)
Input
3. Keith
Gates
b. Ernest
Max
c. Jefir
Chip

11

11

1

1

The Perceptual

Level

[khi.s] —>
[kej.ts”] —>
[ar.ni.sth] —>
[mekhs] —)
[tcef] —)
[tshiphl —>

The Operative Level
[tchisi]
[kaj.tsi]

[e.ni.si.t"a]
[maj.khasi]

[tcefu]
[tshi-phu}

According to Lin, the epenthetic nucleus is realized as [i] through coronal

assimilation if preceded by a coronal sibilant, as [u] through labial assimilation if

preceded by a labial consonant, and as [a] by a default rule if preceded by another

non-labial consonant. As I have shown in §3.3, if the acoustic signal of English [j] is

78

 

represented as [cw] and it is assigned to the coda position, then the following

epenthetic nucleus is realized as [y]. Note that the non-contrastive feature [labial] may

be deleted during the Operative Level for the 'character selection' reason, and the

result is that the epenthetic nucleus surfaces as [i]. Some examples are shown in (70):

(70)
Input

a. Cash
Nash

b. Bush
Page

1 1

1

1

The

Level

Ik"et"’l
[neswl

[pusw]
[phejtsw]

Perceptual

3.5.3 Liquids in the Coda Position

1

1

1

1

The Operative Level L

 

.qw.'.l-'.. -

[k’aj-syl
[noisy]

[pusa]
[phei-tsil

Liquids are not legal coda consonants in Mandarin. When English forms enter

Mandarin, the liquids that are assigned to the coda position are either preserved by

vowel epenthesis or simply deleted. In this section, I will (i) discuss what causes this

asymmetry and (ii) find out the relevant constraints and give a plausible analysis.

Before we reconsider the data, there is one question that we need to answer. What is

the reason that in the loanword adaptation liquids may be deleted while obstruents are

always preserved? Following Yip (1993), I suggest that this asymmetry be attributed

to the difference in salience between liquids and obstruents. First of all, compare the

79

English form-initial obstruents with the immediately following liquids in the

Cantonese examples given in (71):

(71) Cantonese loanword adaptation (Silverman 1992: p.290, Yip 1993: p.267, 270)

a. b.

Input Output Input Output
brake ——> [pik.lik] broker —) [puk.ka]
cream ——) [kej.lim] freezer —> [fi.sa]
plum —> [pow.lam] place —-> [phej.si]
fluke —) [fu.luk] floorshow —> [fo.sow]

We see that the obstruents are all preserved but the liquids are either preserved
(71a) or deleted (71b). Silverman (1992: p.290) suggests that the liquids in (71a) are
preserved in order for the output to be bisyllabic; otherwise, the liquids are deleted.
Following Silverman’s idea, Yip (1993: p.268) further explains why the liquids are
deleted while the obstruents are not. She suggests ‘that the liquids are less salient than
the preceding obstruents and that this lack of salience renders them relatively
vulnerable to deletion.’ To summarize, since liquids are unsalient by nature, which in
turn leads to their vulnerability to deletion, whenever the output forms are guaranteed
to be at least bisyllabic, the liquids undergo deletion (e.g., the output forms in (71b)

would satisfy this constraint even though the liquids are deleted).

80

 

 

In addition, it has been widely stated in the literature that liquids are vowel-like
and thus sometimes not parsed. For example, Fay & Culter (1977) claim that liquids
possess vowel-like formants so that they fail to saliently stand out from surrounding
vowels. Zue (1985: p.141) also suggests that liquids are similar to vowels in terms of
formant position. Since liquids are unsalient and have difficulties in standing out from
neighboring vowels, it makes sense that they tend to be deleted in loanword
adaptation.

Nevertheless, our data show that the liquids assigned to the onset position are
almost always preserved”; only those assigned to the coda position may be deleted.
This might be because our data basically comprise adaptations of proper names. It is
reasonable that when it comes to proper names, people tend to preserve more
segments, even if they are not that salient. Therefore, though liquids are unsalient in
general, in proper name transliteration, onset liquids are still considered salient
enough to be preserved; only some coda liquids that are really unsalient must go. This
issue, however, needs further research and I leave it open here.

Let us now move on to our data. Consider the examples given in (72) and (73):

 

‘2 The only case 1 have found where an onset liquid is deleted is Fgaser —> [fej.c"ej].

81

 

(72)
Input

Pierce
Clare
Blair
Sears

e-mail

(73)
Input

Ann Arbor
Mark
Barbara

Starr
Mort

11111

11111

Output

[phiersi]
[khalajor]
[pu.lajar]
[ci.er.si]

[i.mej.er]

Output

[an.na.paw]
[makho]
[pa.pa.|a]
[§i.tha]
[moths]

Input

Malcolm
Nobel
Miles

Neilson
Hillman

Input

Orson
Gordon
Goldberg
Nicole
Harold

11111

11111

Output

[maj .er.khon]
[nwo.pej.er]
[maj.er.si]
[ni.er.sen]
[ci.er.man]

Output

[c.sen] '
[keten] t-
[ke.po]

[ni.k”e]

[xa.lwo.te]

 

In (72), the English coda liquid is preserved with a schwa epenthesized between

it and the preceding vowel, while in (73) the English coda liquid is deleted. As

mentioned in §1.4.3, Mandarin only allows [n] and [1]] in the coda position. However,

the examples in (72) show that [r] is able to occur in this particular position as well. In

fact, there are two circumstances in Mandarin native phonology in which [r] can serve

as the coda: the special syllable [or] and [r]-suffixed words. Some examples for each

type are given in (74):

82

(74) Two types of Mandarin native words with [r] in the coda position (the numbers
denote the tones)

(a) The underlying type (b) The derived type
[or]2 ‘son’ [xwa] 1 + [r] —> [xwar]1 ‘flower’
[or]3 ‘ear’ [pu]4 + [r] —> [pur]4 ‘cloth’
[ar]4 ‘two’ [wo] 1 + [r] —> [wor]1 ‘nest’
“(9] 1 + If] —-> [kor]1 ‘song’

The reason I call [or] a ‘special’ syllable is because it is the only Mandarin

syllable with [r] in the coda position. Given the fact that [or] is well formed, I assume

that [r] is a legal coda consonant as long as it occurs in the syllable [er]. Derived

syllables such as those in (74b) are considered ill-formed in Mandarin loanword

adaptation for two reasons. First, while [r]-suffixed words are extensively found in

Beijing Mandarin, they are not common in Standard Chinese (Duanmu 2000: §6),

which is the target language in this thesis and referred to as Mandarin simply for

convenience. Second, the suffix [r] may add the meaning of ‘smallness’ to the word to

which it attaches (Duanmu 2000: p.195), but we have never found any English

loanwords in Mandarin that semantically involve ‘smallness’.

It is therefore necessary to modify the definition of the constraint CODA

CONDITION. The modified version is given in (75):

(75) CODA CONDITION: Only [n] or [1]] is allowed in the coda position. [r] is allowed

in this position only if it occurs in the syllable [or].

83

 

We now turn to the central question of this section: What is the reason that in
(72) the English coda liquids are preserved while in (73) they are deleted? To answer
this question, we need to consider the contexts in which they occur. We find that in
(72) the English coda liquids are preceded by [i]/[i], [e], [a] or [I]. In (73), however,
we do not find any of these vowels preceding the liquids; instead, they are preceded
by [a], [c] or [e]. If we further consider the qualities of these two groups of vowels,
we find that only front/non-front can distinguish one group from the other. It therefore
seems reasonable to assume that the preservation-deletion asymmetry in the English
coda liquids is attributed to whether the preceding vowel is front or not. I suggest that
if the English coda liquid is preceded by a front vowel, it becomes salient and must be
preserved; on the other hand, if it is preceded by a back vowel, it is very unsalient and
must go.l3

Whether a segment is salient or not sometimes depends on the context in which it
occurs. The reason a liquid is usually less salient than an obstruent is that it is
vowel-like and thus not able to saliently stand out from its neighboring vowel(s).
However, we do find a context in which a liquid is perceived as salient. In English, a
sequence such as a front tense vowel plus a liquid within a syllable results in the

perception of an intervening schwa; for example, eel is perceived as [i°l] and air as

 

'3 A few exceptions are found. For example, Darwin surfaces as [ta.er.wen] rather than as *[ta.wan].
This might be because of the influence of the spelling. In general, English coda liquids preceded by
back vowels are deleted in the adaptation.

84

 

 

 

 

n.” ,l

 

 

[ej°r]. The reason for perceiving an intervening schwa in such a context is simple: To
pass from a front tense vowel to post-alveolar/retroflexed [r] or velarized [1] requires
considerable movement of the tongue; if this is somewhat slowed, an epenthetic
schwa develops as the tongue moves through schwa space (Gick & Wilson 2001: p.1).
Because an intervening schwa is perceived and considered part of the liquid, the
liquid becomes quite salient.

This phonetic phenomenon is found in [r]-suffixed words in Beijing Mandarin as

 

well. It is Observed that if the suffix [r] directly attaches to the vowel [i] or [y], an

intervening schwa is perceived. Some examples are given in (76):

(76)

[ti]4 + [r] ——> [tjor]4 ‘younger brother’
on + [r] —> [uar]2 ‘fish’

[phiIZ + [r] —> [phjer]2 ‘skin’

Because [i] and [y] are incompatible with retroflexion in articulation, when an
Open syllable with one of these vowels is suffixed by [r], a schwa is epenthesized
between the vowel and the [r] in order to separate one from the other (Chao 1968:
p.46, c. Cheng 1973: p.25 and Lin 1989: p. 102).

The fact that an English sequence of a front tense vowel plus a coda liquid

requires considerable movement of the tongue actually implies the incompatibility of

85

the two segments. Since they are incompatible with each other, the liquid, though
believed to be unsalient by nature, is easily noticed and thus must be salient.

Because a front tense vowel-liquid sequence in English results in the perception
of an intervening schwa, the schwas Observed in [phi.ar.si] ‘Pierce’, [ci.ar.si] ‘Sears’,
[i.mej.or] ‘e—mail’, [maj.ersi] ‘Miles’ and [ni.ar.sen] ‘Neilson’ in (72) should not be
treated as epenthetic. Instead, they should be thought of as appearing at the Perceptual
Level and being preserved at the Operative Level of the adaptation. Those schwas
observed in the remaining examples are truly epenthetic, given that the preceding
vowels in the English forms are lax rather than tense.l4 For the sake of clarity,

compare the adaptations in (773) with those in (77b):

(77a)

Input The Perceptual The Operative Level
Level

Clare —-> [khler] —> [Ithalajar]

Malcolm ——> [mel.kham] —~> [maj.grkhen]

Nobel —> [no.pel] -—> [nwo.pej.§r]

Hillman —> [xil.man] —> [cinnman]

 

’4 I assume that an English coda liquid preceded by a front lax vowel is salient as well because the
tongue still has to move quite a distance.

86

 

(77b)

Input The Perceptual The Operative Level
Level

Pierce —> [phi.ar.s] —> [phi.ar.si]

Blair —> [pu.le.er] —> [pu.lajer]

Sears —> [si.er.s] —-> [ci.er.si]

e-mail —> [i.mej.el] —> [i.mej.ar]

Miles —) [maj.als] —> [maj.ar.si]

Neilson —-) [ni.al.sen] —> [ni.er.sen]

In (77a), the underlined schwas at the Operative Level are epenthetic because
they lack a correspondent at the Perceptual Level. In (77b), the input forms all possess
a front tense vowel-liquid sequence. As a result, a schwa is perceived and appears at
the Perceptual Level, which subsequently shows up at the Operative Level.

A question now arises: If the schwa is epenthetic, why is it always epenthesized
before the liquid? Put differently, why do we never find a case where a schwa is
epenthesized after the liquid (both [re] and [19] are well-formed Mandarin syllables),
just like the way an obstruent coda is fixed (e.g., Matt —> [maj.thaD? I suggest that the

faithfulness constraint SROLE plays a role here:

(78) SROLE-IO (Gafos 1998: p.227): A segment in the input and its correspondent in

the output have identical syllabic roles.

To illustrate how this constraint works, compare the forms at the Perceptual
Level with those at the Operative Level in (79):

87

 

(79)

Input The Perceptual The Operative Level
Level

Clare —+ [khler] —> [khalajar] *[khe.laj.re]

Nobel ——> [no.pel] —-> [nwo.pej.er] *[nwo.pej.le]

At the Perceptual Level, the liquids in question are in the rime. If a schwa is

epenthesized before them, the liquids stay in the rime at the Operative Level,

satisfying SROLE-IO. In contrast, if a schwa is epenthesized after the liquids, the

liquids are reassigned to the onset position at the Operative Level, violating SROLE-IO.

Since [or] is a well-formed Mandarin syllable, fixing a problematic liquid coda by

epenthesizing a schwa before it simultaneously satisfies both CODA CONDITION and

SROLE-IO. On the other hand, if a schwa is epenthesized after the liquid coda, as

shown in the starred forms, only CODA CONDITION is satisfied.

As for the ranking of SROLE-IO, I claim that it be ranked below CODA

CONDITION.

(80) CODA CONDITION >> SROLE-IO

 

 

 

 

 

 

(81)
khler ‘Clare’ CODA CONDITION SROLE-IO DEP-V-IO
a. khalajr *(F)! g . _ 7 . 7’19). ,
CE’b. khalajor *(o)*(e)
c. khe.laj.ra *(r)! *(o)*(o)

 

 

 

 

 

88

Candidate (3) is ruled out first because it violates the undominated constraint
CODA CONDITION. Candidate (c) later loses to Candidate (b) because Candidate (b) is
faithful to the input in terms of the syllabic role of the liquid [r] while Candidate (c) is
not. Note that in all the three candidates, neither [kh] nor [1] violates SROLE-IO in spite
the fact that a schwa is epenthesized between them.

The motivation for ranking SROLE-IO below CODA CONDITION comes from the

cases in which the coda is an obstruent. Consider the tableau in (82):

 

 

 

 

 

(82)
meth ‘Matt’ CODA CONDITION SROLE-IO DEP-V-IO
G’a. maj.t”a * I ’ 7 '*(’a)
b. majth *(th)! ,
c. majeth *0")! .’_"(e)

 

 

 

 

 

Although Candidate (a) violates SROLE-IO, it is still selected over the other

candidates because it is the only one that satisfies CODA CONDITION. This tableau

shows that whenever the input has an obstruent coda, the optimal candidate must

violate SROLE-IO because an obstruent coda is always fixed by epenthesizing a vowel

after it.

Two things are worth mentioning here. First, SROLE-IO does not seem to interact

with other constraints and thus cannot be ranked with them. For example, SROLE-IO

and DEP-V-IO cannot be ranked with respect to each other because a vowel is never

89

 

m-mi- .1.- -. t». as.” - .
‘l'
I

epenthesized in order to satisfy SROLE—IO. or vice versa. Second, the effect of

SROLE-IO is usually masked by the undominated constraint CODA CONDITION. For

example, an obstruent coda is always fixed by epenthesizing a vowel after it (thus

violating SROLE-IO) since an obstruent is never allowed in the coda position. The

only situation in which SROLE-IO takes effect is when the input possesses a

problematic liquid coda. In such a situation, the liquid coda can be fixed by

epenthesizing a schwa either before or after it. Since adding a schwa before the liquid

does not change the liquid’s syllabic role while adding a schwa after the liquid does,

the former is favored.

Let us now turn to the claim that English coda liquids preceded by a back vowel

are unsalient and must be deleted in the loanword adaptation. As I have stated above,

whether a segment is salient or not may depend on the context in which it occurs. In

an English syllable where a coda liquid follows a front vowel, the liquid is salient

because the articulation of such a sequence requires considerable movement of the

tongue (that is, the two segments are incompatible with each other in terms of

articulation). We can therefore assume that an English coda liquid preceded by a back

vowel should be unsalient given that to pass from a back vowel to

post-alveolar/retroflexed [r] or velarized [I] does not involve too much movement of

the tongue (they already have a similar tongue body position). Indirect evidence for

90

 

this assumption is available from the speech errors made by Chinese speakers who are
learning English. Gui (1985) reports that when Chinese speakers learn English, they
tend to delete the coda [I] that is preceded by [o], [3] or [‘3]. Gui’s data also show that
coda [r] may be deleted if it is preceded by [a] and followed by [n]. Note that [o], [o],
[53] and [a] are all back vowels. The close match between what is deleted in Mandarin
loanword adaptation and what tends to be deleted by Chinese speakers in their speech
errors seems to suggest that the two processes are triggered by the same mechanism.
If we further assume that Chinese speakers choose to delete the most unsalient
segments when they are under pressure and forced to simplify the syllable structure of
the English form (as opposed to obstruents in onset clusters, where vowel epenthesis
is found), then the claim that unsalient segments undergo deletion in Mandarin
loanword adaptation is plausible.

Steriade (2001) proposes that the ability of speakers to adapt a non-native string
to the closest equivalent in their native inventory has a perceptual basis. This idea is

formulated as in (83):

(83) Perceptual similarity to input (Steriade 2001: p.222)

The likelihood that a lexical representation R will be realized as modified R’ is a

function of the perceived similarity between R and R’.

91

 

r.

I ”‘1’ . '.._.'._

I?

She states that ‘certain contrasts are more discriminable than others, and that the
same contrast is more salient in some positions than in others (p.236).’ To illustrate
this idea, compare the adaptation of tips with that of post when they enter Cantonese,
tips -—) [thip.si] and post ——> [phosi]. We observe that the sibilant [s] is preserved in the
coda position while the stop [t] is not. This asymmetry is of course attributed to the
nature of the two segments. In other words, in the coda position, the contrast of [s] —)
G is more discriminable than that of [t] -—> E because [5] is by nature more salient
than [t]. We also observe that the stop [t] is not deleted in all contexts. It is deleted
only if it is syllable-final and preceded by another consonant. In the onset position, it
is preserved. The preservation-deletion asymmetry of [t] is obviously dependent on
the context in which it occurs; namely, the contrast of [t] —> E is more easily noticed
in the onset position than in the coda position.

In an OT analysis, we can explain it by treating MAX-C(strident) as undominated
(e.g., [s] is never deleted in any position) but ranking MAX-C(stop)/C_# below DEP-V
(e.g., in post ——) [pho.si], the problematic form-final [t] is deleted rather than being
preserved by vowel epenthesis). The different rankings of various MAX-C constraints
indicates that the contrast of sibilant —> E and that of stop —-) ®/C__# have different
similarity values. The different similarity values in turn reflect the difference in

auditory salience among the various C —> g contrasts for Cantonese speakers.

 

The idea that various C —) Q contrasts have different similarity values can help
us explain the fact that in the loanword adaptation, coda liquids are either preserved or
deleted, and obstruents are always preserved. Based on the observation that the only
consonants found to be deleted in the adaptation are those coda liquids preceded by
back vowels, I suggest that for Mandarin speakers the contrast of liquid —> r

®N[+back]_C0] has a greater similarity value than the contrasts of obstruent —> E and

 

liquid ——> @N[—back1_Co] have. Similar to the Cantonese example, the MAX-C-IO

 

constraint corresponding to the contrast of liquid -—) @leackLCo] is ranked below

DEP-C-IO while the MAX-C-IO constraints corresponding to the contrasts of obstruent

-—) Q and liquid ——) ®N[_mck]_C0] are ranked undominated. Consider the tableau

given in (87), where [markh] ‘Mark’ serves as the input:

(84) MAX-C-IO(obstruent): Input obstruents must have output correspondents.

(85) MAX-C-IO(|iquid/V[-back1_Co): Input liquids that are preceded by front vowels

must have output correspondents.

(86) MAX-C-IO(liquid/V[+back]_Co): Input liquids that are preceded by back vowels

must have output correspondents.

93

(87)

 

 

 

 

 

 

markh ‘Mark’ CODACOND MAX-C-IO MAX-C-IO DEP-V-IO MAX-C-IO
(obstruent) (liquid! (liquid/
V[—back]_C0]) v[+back]_CO])
G'a. ma.kh9 *(9) *(l’)
b. ma.ar.kh9 *(9)*(9)!
c. mankha *(f)! *(9)
d. ma *(kh)! *(r)

 

 

 

 

 

 

 

Candidate (c) and Candidate (d) are ruled out in the very beginning because they

violate the undominated constraints CODACONDITION and MAX-C-IO(obstruent),

respectively. Candidate (b) later loses to Candidate (a) because Candidate (b) chooses

schwa epenthesis as the strategy to fix the problematic coda [r] and violates the more

serious constraint DEP-V-IO while Candidate (a) simply deletes the [r], violating the

least significant constraint MAX-C-IO(Iiquid/V[+back]_Co]).

Now consider another tableau. where the input possesses a sequence of a front

vowel plus a coda liquid:

 

 

 

 

 

(88)
khler ‘Clare’ CODACOND MAX-C-IO MAX-C-IO map-v.10 MAX-C-IO
(obstruent) (liquid/ (liquid/
Vl—back]_C0]) V[+backl..CO])
ea. kha.laj.ar *(9)*(9)
b. khalaj *(r)! *(9)
c. k"9.lajr *(r)! *(a)

 

 

 

 

 

 

 

94

 

Since the problematic coda [r] is preceded by a front vowel, it is perceived as

salient and therefore cannot be deleted. The only way of fixing it is by schwa

epenthesis. Candidate (:1) wins over its competitors because it is the only one that

adopts this strategy. The other two candidates either delete the [r] (Candidate (b)) or

simply leave it alone (Candidate (c)), violating the undominated constraints

MAx-C-IO(Iiquid/V[_back]_C0]) and CODACONDITION.

3.5.4 [In] in the Coda Position

3.5.4.1 Polysyllabic Input

Though Mandarin has the bilabial nasal [m], it is observed in our data that

English coda [m] surfaces as [n] or [I3] in a vast majority of cases. It is apparent why

this happens: Mandarin does not allow [m] in the coda position. Following Silverman

(1992: p.297), I assume that the acoustic signal of English [m] is perceived by

Mandarin speakers as [m] no matter in what position in the input it occurs, and once

the adaptation arrives at the Operative Level, where Mandarin syllable structure

constraints hold, those that are assigned to the coda position are fixed. It is interesting

that Mandarin fixes coda [m] by turning it into [n] or [1]] instead of vowel epenthesis,

which, as we have seen in the previous sections, is used to fix coda obstruents and

liquids that are preceded by front vowels. This indicates that DEP-V-IO outranks

95

 

IDENT—IO(place). The tableau in (9]) shows that for the input [mi.ri.am] ‘Miriam’, this

ranking surfaces [mi.rwej.an] instead of [mi.rwej.a.mu].

(89) MAX-C-IO(nasal): Input nasals must have output correspondents.

 

 

 

 

 

 

 

.,....
(90) IDENT-IO(place): The specification for place of articulation of an input segment
must be preserved in its output correspondents.
(91) 1'!
miriom ‘Miriam’ CODACOND MAX-C-IO(nasal) DEP-V-IO IDENT—IO(place) E

G'a. mi.rwej.an *(Lab -> Cor)

b. mi.rwej.am *(m)! L

c. mi.rwej.a *(m)!

d. mi.r‘"ej.a.mu *(u)!

 

 

 

 

 

 

Candidate (b) and Candidate (c) are ruled out because the former keeps the [m]

in the coda position, violating CODA CONDITION, and the latter deletes the [m],

violating MAX-C-IO(nasal). The remaining candidates then move on for the

evaluation of DEP-V—IO and IDENT—IO(place). Candidate (3) is selected over Candidate

(d) as the output because Candidate (a) violates the lower-ranked constraint

IDENT-IO(place) by turning the coda [m] into [n] while Candidate (d) fixes this

ill-formed structure by vowel epenthesis, violating the more significant constraint

DEP-V-IO.

The above analysis seems to imply that to fix an illicit coda, a feature change

96

costs less than vowel epenthesis and thus should be preferred. However, we have

never found an example in our data where a perceived obstruent or liquid assigned to

the coda position undergoes a feature change and turns into a nasal. I suggest that this

can be explained by ranking IDENT-C-IO(nasal) above DEP-V-IO. Consider the tableau

in (93), where for the input [sikrit] ‘Sigrid’, [ci.k9.li.ta] is selected over [ci.ka.lin] as

the output:

(92) IDENT-IO(nasal): Correspondent segments in input and output have identical
values for the feature [nasal].

 

 

 

 

 

 

(93)

sikrit ‘Sigrid’ CODACOND MAX-C-IO(Obstruent) IDENT-C-IO(naS) DEP-V-IO
a. ci.k9.lit *(t)! *(9)  

9b. ci.ka.li.to *(9)*(9)
c. 9i.ka.lin *(t —> n)! *(a)
d. ci.ka.li *(t)! . *(9)

 

 

 

 

 

 

Candidate (a) and Candidate ((1) are ruled out because Candidate (a) violates

CODA CONDITION by keeping the [t] in the coda position and Candidate violates

MAX-C-IO(obstruent) by deleting it. Candidate (c) loses to Candidate (b) because

Candidate (c) turns the [t] into [n] in response to the requirement of CODA CONDITION

and the accompanying violation of IDENT—IO(nas) is more serious than the violation of

DEP-V-IO incurred by Candidate (b).

97

 

 

It should be noted here that IDENT-C-IO(nas) is ranked undominated. The reason
is that we find no case in the data where an oral consonant has turned into a nasal, or
vice versa. In other words, IDENT—C-IO(nas) is ranked undominated because changes
of the values for [nasal] in consonants never happen.

Here is another example showing that IDENT—C-IO(nas) is undominated. In this

example, the coda [r] is not fixed by being turned into [n].

 

 

 

 

 

 

(94)
khler ‘Clare’ CODACOND MAX-C-IO IDENT-C-IO(nas) DEP-V-IO
(liquid/V [_back1_C0])

a. khalajr *(r)! ' *(o)
6’b. k“a.laj.ar *(a)*(a)
c. khalan *(r —) n)! *(a) I
d. khaJaj *(r)! *(a).

 

 

 

 

 

 

3.5.4.2 Monosyllabic Input

Not every perceived [m] assigned to the coda position turns into [n] or [1]],
though. We have found in our data that coda [m] sometimes does not undergo a
feature change but is retained with a vowel epenthesized after it, just like the way an

obstruent coda is fixed. Some examples are given in (95):

98

 

(95)

Input The Perceptual The Operative Level
Level

77m —> [thim] —9 [mm]

Jim —> [tcim] —> [toim]

Tom —> [tham] —> [than-m]

Sam —> [sem —> [sanm]

In (95), the perceived [m]s are assigned to the coda position. However, unlike

what we have just discussed, they surface intact with the vowel [u] epenthesized after

them and occupy the onset positions of the newly formed syllables. If we look at the

number of the syllables in both the perceived and surface forms, we find that all the

perceived forms are monosyllabic and all the surface forms are bisyllabic. It therefore

seems that whether or not coda [m] is fixed by vowel epenthesis is related to whether

or not the input is monosyllabic. Broselow, Chen & Wang (1998) analyze the

simplification of English forms with obstruent coda by Mandarin speakers learning

English as a second language and discover that they tend to add an extra vowel at the

end of a monosyllabic word but delete the obstruent coda if the word is bisyllabic. For

example, /VIg/ —> [v1.99] while /filIg/ —> [fi.lI]. They claim that the preference for

bisyllabic words is playing a role here and that it represents a case of the emergence

of the unmarked, which means that this preference, though not obviously motivated,

exists in the grammar of Mandarin. I agree with them in that in Mandarin native

99

 

phonology, this preference is masked by higher ranked faithfulness constraints such as
MAX-C-IO and DEP—V-IO and thus has no effect (for example, the underlying form
/khan/ ‘to see’ surfaces as [khan] rather than *[kha.n9]); nevertheless, this preference
exposes itself whenever forms with forbidden coda serve as the input, e.g., foreign
forms. In such circumstances, since the faithfulness constraints must be violated in
order to satisfy Mandarin phonotactics, the preference for bisyllabic words comes into

force. The tableau in (97) illustrates their idea:

(96) WD BIN: Word should consist of two syllables (Broselow, Chen & Wang (1998:
p.272)).

 

 

 

 

 

(97) (p.273)
/VIg/ CODA CONDITION DEP-V-IO, MAx-C-IO WD BIN
a. VIg *! V l *.
G‘“’b. v1.99 *
c. v1 * *!

 

 

 

 

Candidate (a) is ruled out because it violates the undominated constraint CODA

CONDITION by possessing [g] in the coda position. The other two candidates, which

both satisfy CODA CONDITION, then move into the second round where they are

evaluated by the faithfulness constraints banning vowel epenthesis and consonant

deletion. Since they tie in this round, the decision is up to the lowest ranked constraint

WD BIN. Candidate (b) finally wins out because it possesses two syllables.

 

 

In addition to Broselow, Chen & Wang, Duanmu (2000: §7) claims that the
majority of commonly used vocabulary words in Mandarin are bisyllabic and there is
evidence that bisyllabic words are still increasing even today. Li & Thompson (198])
also claim that ‘although classical Chinese appears to have been a monosyllabic
language, modern Mandarin is no longer monosyllabic; instead, Mandarin has a very
large number of polysyllabic words (p. l4).’ Yip (1993) observes a similar tendency in
Cantonese — ‘kinship names and hypocoristics are subject to a bisyllabic minimum,
and since most names are monosyllabic, they are augmented by the addition of the
prefix [a], e.g., /jip/ —> [a jip] (p.276)’ — and suggests the constraint Minimal Word,
which requires each word to consist of at least two syllables. Following Broselow,
Chen & Wang, I assume that the vowel epentheses in (95) are triggered by the
preference for bisyllabic words and will use the constraint WD BIN in the subsequent
analysis.

Nevertheless, our case is more complicated than the one analyzed by Broselow,
Chen & Wang. Recall that in their analysis, the two faithfulness constraints
MAX-C-IO and DEP-V—lO are ranked above WD BIN but not ranked with respect to
each other. This ranking entails that consonant deletion and vowel epenthesis cost the
same, and since input forms ending in obstruents cannot surface faithfully anyway (in

other words, the obstruent coda must be fixed by either vowel epenthesis or deletion),

lOl

 

 

the alteration proceeds in the direction of satisfying WD BIN. In our case, the
trouble-making coda is not an obstruent but [m], which presumably may undergo
deletion, vowel epenthesis or feature change to satisfy CODA CONDITION. However,
we cannot simply rank WD BIN below MAX-C-IO, DEP-V-IO and IDENT-IO(place),
since by doing so, we would get an unwanted output. Consider the tableau in (98),

where [thim] ‘Tim’ serves as the input:

 

 

 

 

 

 

(98)
thim ‘Tim’ CODACOND MAX-C-IO(nasal) DEP-V-IO IDENT-IO(place) WD BIN
a. thim *! v 7 l 8*, 7
anthrmu *(u)! I ,. .
CFC. thin *(m -—> 13) *
d. thi *! ' j *

 

 

 

 

 

 

 

Candidate (a) is ruled out because of its fatal violation of CODA CONDITION. The
other three candidates, which all have done something to the coda and thus satisfy
CODA CONDITION, stay in the contest competing for being the output. Candidate (c)
finally surfaces as the output because the constraint it violates by fixing the coda
(IDENT-IO(place)) is less serious than those Candidate (b) and Candidate ((1) violate
(DEP-IO and MAX-IO(nasal), respectively). The attested output, however, is
Candidate (b) rather than Candidate (c), as indicated by the sad face.

The problem with this ranking is that the effect of WD BIN is still masked. This is

102

am

 

because we can determine the output simply by the different rankings of
MAX-C-IO(nasal), DEP-V-IO and IDENT-IO(place) without considering WD BIN, as
indicated by the entire shading of the WD BIN column in (95). To put it differently,
the output is always the candidate that changes the [m]’s place of articulation in
response to the requirement of CODA CONDITION. By promoting WD BIN to a higher r

position in the constraint hierarchy, WD BIN can be activated. I claim that WD BIN be

 

ranked below MAX-C-IO(nasal) but above DEP-V-IO.

l‘

(99) MAX-C-IO(nasal) >> WD BIN >> DEP-V-IO

Consider the tableau in (100), where the rankings of the constraints have been

 

 

 

 

 

rearranged:
(100)
thim ‘Tim’ CODACOND MAX-C-IO(nasal) WD BIN DEP-V-IO IDENT-IO(place)
a. thim *(m)! *
C?b. thimu *(U)
c. thin *! *(m —> 1])
d. thi *(m)! *

 

 

 

 

 

 

 

After promoting WD BIN from the bottom to the current position, we see that
Candidate (b) is selected as the output. Candidate (a) and Candidate (d) are still the

worst candidates, since they violate the undominated constraint CODA CONDITION and

103

MAX-C-IO(nasal), respectively. Candidate (c) loses to Candidate (b) due to its more

serious violation of WD BIN.

3.5.4.3 An Attempt to Account for Cases such as Lynn

I have argued in the previous section that the preference for bisyllabic words

triggers the vowel epentheses observed in (92) and that the corresponding constraint

WD BIN should be ranked above DEP-V-IO. However, we have found some

counterexamples indicating that WD BIN has to be ranked below DEP-V-IO. Consider

the adaptations given in (101):

(101 )

Input Output Input Output

Ben —> [pan]/[pan.an] King —9 [tcin]/[tcin.an]
Dan —-> [tan]/[tan.9n] Lynn -> [lin]/[linen]
Ken —> [khan]/[khan.an]

In (101), each English form has two Mandarin equivalents, one monosyllabic

and the other bisyllabic. I have argued in §3. 1.1 that for the perceived forms to surface

as bisyllabic, [a] is epenthesized right before the [n], and the feature [nasal] involved

in the vowels surfaces as [n] due to the requirement that a full Mandarin syllable is

bimoraic. The constraint ranking I have just argued for has no trouble surfacing the

104

 

bisyllabic forms. Take a look at the tableau in (102), where for the input [lin], the

bisyllabic candidate [linen] is selected as the output over the monosyllabic one [lin]:

 

 

 

 

(102)

lin ‘Lynn’ CODA COND MAx-C-IO WD BIN DEP-V-IO IDENT-IO(place)

a.lin *! __ I ' 7 i r...
9b. lin.9n *(9)

 

 

 

 

 

 

 

 

The problem we are faced with now is that, based on this particular ranking,

1'-.l.:='lc -lll - . ' . I

there is no way in which the English forms in (101) can surface as monosyllabic:

since WD BIN is ranked above DEP-V-IO, whenever a monosyllabic form serves as

the input, it must be augmented to bisyllabicity at the expense of violating DEP-V-IO.

This problem forces us to reconsider the constraint ranking. The fact that in (101)

the output forms may be monosyllabic seems to imply that at least for some Mandarin

speakers, satisfying WD BIN is not really that important. As a result, WD BIN should

be demoted in the constraint hierarchy. I suggest that for the examples in (101),

DEP-V-IO and WD BIN are not ranked with respect to each other.

(103) WD BIN >> DEP-V-IO; DEP-V-IO >> WD BIN

In (104), the constraint ranking is the same as that in (102) except that WD BIN is not

ranked with respect to DEP-V-IO, and we can see that both the monosyllabic

105

candidate [lin] and the bisyllabic candidate [linen] surface as the outputs:

 

 

 

 

(104)

lin ‘Lynn’ CODA COND MAX-C-IO DEP-V—IO WD BIN IDENT-I0(place)
CE'a. lin *(9)
9b. lin.an *

 

 

 

 

 

 

It should be noted here that this change does not affect the analysis done above.

Take a look at the following tableau, which is the same as the one given in (104)

except for the relevant rankings of DEP-V-IO and WD BIN:

(105)

 

thim ‘Tim’

CODA COND

MAX-C-IO

WD BIN

DEP-V-IO

IDENT-IO(place)

 

a. thim

*1

 

c3Tb. thi.mu

 

c. thiN

*(m 9N)!

 

 

d. thi

 

 

 

*1

 

 

 

We see that Candidate (b) is still selected as the output after the change to the

ranking is made.

106

 

 

 

CHAPTER 3

CONCLUSION

In this thesis, I argue that English loanword input to Mandarin is best viewed as

consisting of superficial acoustic signals. My argument is based on the observation

that English non-contrastive features may appear in the output of the loanword

adaptation and that two English segments in an onset cluster may merge into one.

Paradis & Lacharité’s (1997) claim that loanwords enter the borrowing language with

structures cannot account for the second observation, since it is unable to explain why

the complex onset is not repaired by vowel epenthesis.

I also argue in this thesis that when English forms enter Mandarin, the acoustic

signals are interpreted using the allophones of Mandarin. This is also in contradiction

with Paradis and Lacharité’s claim that loanword input enters into the borrowing

language’s lexicon. This fact suggests that although the Operative Level of the

loanword adaptation and the lexical level of Mandarin have the same function, their

inputs are composed of segments from different sources.

Finally, I argue that the preservation-deletion asymmetry found in the coda

liquids is attributed to the difference in salience. I suggest that though liquids are

unsalient in nature, English coda liquids that are preceded by front vowels are

107

 

 

perceived as salient and thus must be preserved in Mandarin loanword adaptation.

Some issues are left open in this thesis and deserve attention in the future. One is

regarding the fact that some types of feature changes never occur in the loanword

adaptation. For example, we do not find the feature change [+continuant] <—)

[—continuant]. This seems to suggest that these feature changes make greater

difference in perception than others and therefore are not tolerated in the loanword

adaptation. The other issue has to do with the observation that only the liquids

assigned to the coda position may be deleted while those assigned to the onset

position are almost always preserved. Although this might be attributed to the nature

of our data, the syllabic position to which they are assigned is probably playing a role

as well.

108

 

F’h ’ '. - A.
;
d
'1

REFERENCES

Bao, Zhiming. 1996. The Syllable in Chinese. Journal of Chinese Linguistics 24:
312-354.

Broselow, Ellen, Chen, Su-I and Wang, Chilin. 1998. The Emergence of the
Unmarked in Second Language Phonology. Studies in Second Language
Acquisition 20: 261-280.

Chao, Yuen-ren. 1968. A Grammar of Spoken Chinese, Berkeley and Los Angeles,

University of California Press.

Cheng, Chin-chuan. 1973. A Sync/Ironic Phonology of Mandarin Chinese. Mouton,
The Hague.

Duanmu, San. 2000. The Phonology of Standard Chinese. Oxford, New York: Oxford

University Press.

Fay, David and Culter, Anne. 1977. Malapropisms and the Structure of the Mental
Lexicon. Linguistic Inquiry 8: 505 — 520.

Gafos, Diamandis. 1998. Eliminating Long-Distance Consonantal Spreading. Natural
Language & Linguistic Theory 17: 223 - 278.

Gick, B. and Wilson, I. 2001. Pre-liquid Excrescent Schwa: What Happens When
Vocalic Targets Conflict. In P. Dalsgaard, B. Lindberg and H. Benner (eds.)
Proceedings of the 7th European Conference on Speech Communication and

Technology 273-276. Aalborg, Denmark: Center for Personkommunikation.

Gui, C. K. 1985. Applied American English Phonetics. Shanghai Foreign Language

Education Press.

Guo, Hugo Ling-Yu. 2001. A Constraint-based Analysis to Mandarin Loanwords:
Evidence from Transliterated American State Names and Typhoon Names.
Precedings of the 7m International and the 19:}: National Conferences on
Chinese Phonology 442-462.

Ham's, John. 1994. English Sound Structure. Massachusetts: Blackwell.

109

 

Kager, Rene. 1999. Optimality Theory. New York: Cambridge University Press.

Kenstowicz, Michael. 1994. Phonology in Generative Grammar. Cambridge:
Blackwell.

Ladefoged, Peter. 1982. A course in Phonetics. New York: Harcourt Brace

Jovanovich.

Li, Charles N. and Thompson, Sandra A. (1981). Mandarin Chinese: A Functional

Reference Grammar. Berkeley: University of California Press.

I...

Lin, Yen-Hwei. 1989. Autosegmental Treatment of Segmental Processes in Chinese

Phonology. PhD dissertation, The University of Texas at Austin.

 

Lin, Yen-Hwei. I992. Vocalic Underspecification in Two Mandarin Dialects: A Case
Against Radical Underspecification. Journal of East Asian Lingustics 1:
219-253.

Paradis, Carole and Lacharite, Darlene. 1997. Preservation and Minimality in
Loanword Adaptation. Journal of Linguistics 33: 379-430.

Paradis, Carole and Prunet, Jean-Francois. 2000. Nasal Vowels as Two Segments:
Evidence from Borrowings. Language 76: 324-357.

Prince, Alan and Smolensky, Paul. 1993. Optimality Theory: Constraint Interaction in
Generative Grammar. MA: MIT Press.

Roca, Iggy and Johnson, Wyn. 1999. A Course in Phonology. Oxford, UK; Malden,
Mass: Blackwell Publishers.

Silverman, Daniel. 1992. Multiple Scansions in Loanword Phonology: Evidence from
Cantonese. Phonology 9: 289-328.

Steriade, Donca. 2001. Directional Asymmetries in Place Assimilation, a Perceptual
Account. The Role of Speech Perception in Phonology 219-249.

Yip, Moira. 1993. Cantonese Loanword Phonology and Optimality Theory. Journal of
East Asian Linguistics 2: 261-291.

110

Zue, Victor W. 1985. Notes on Spectrogram Reading. Cambridge: MIT Press.

11]

 

       
     

 

L lES

willy

IIIIIIIIIIIIIIIIIIIII

“Will\lllllllllllllllll

3 1293 02497